Citation
The return of returning

Material Information

Title:
The return of returning the economic genefit of baccalaureate degree completion after stopping out
Alternate title:
The economic benefit of baccalaureate degree completion after stopping out
Creator:
Lane, Patrick David ( author )
Language:
English
Physical Description:
1 electronic file (184 pages) : ;

Subjects

Subjects / Keywords:
College dropouts ( lcsh )
Genre:
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Review:
An emerging strategy in higher education and workforce development policy circles aims to raise local, state, and national degree attainment rates by targeting those who left postsecondary education after earning significant college credits but without completing a degree. This dissertation examines some of the assumptions behind these programs testing whether these “near completers” who return to finish a degree receive a positive economic return compared to those who do not return to finish a degree. Additionally, this research examines whether their outcomes are impacted by the sector (either public, private non-profit, or private for-profit) of the college or university at which they complete their degree. Finally, this study examines whether individual characteristics affect the likelihood that an individual who has stopped out of college will return to complete a degree. Overall, I find that the economic return varies across racial/ethnic background and that not all subgroups earn a positive return from finishing a degree, but returns do not differ by sector. Finally, I find that many of the factors generally associated with increased educational attainment do not appear to have a relationship with the likelihood of finishing a degree.
Thesis:
Thesis (Ph.D) - University of Colorado Denver.
Bibliography:
Includes bibliographic references.
System Details:
System requirements: Adobe Reader.
General Note:
School of Public Affairs
Statement of Responsibility:
by Patrick David Lane.

Record Information

Source Institution:
|University of Colorado Denver
Holding Location:
|Auraria Library
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
945381852 ( OCLC )
ocn945381852
Classification:
LD1193.P86 2015d L36 ( lcc )

Downloads

This item has the following downloads:


Full Text
THE RETURN ON RETURNING: THE ECONOMIC BENEFIT OF
BACCALAUREATE DEGREE COMPLETION AFTER STOPPING OUT
by
PATRICK DAVID LANE
A thesis submitted to the
Faculty of the Graduate School of the
University of Colorado in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
Public Affairs
2015


11
This thesis for the Doctor of Philosophy degree by
Patrick David Lane
has been approved for the
Public Affairs Program
by
Paul Teske, Advisor
Todd Ely, Chair
Kelly Hupfeld
Susan Clarke
Date: 11/15/2015


Ill
Lane, Patrick David (Ph.D., Public Affairs)
The Return on Returning: The Economic Benefit of Baccalaureate Degree Completion
after Stopping Out
Thesis Directed by Professor Paul Teske
ABSTRACT
An emerging strategy in higher education and workforce development policy circles aims
to raise local, state, and national degree attainment rates by targeting those who left
postsecondary education after earning significant college credits but without completing a
degree. This dissertation examines some of the assumptions behind these programs
testing whether these near completers who return to finish a degree receive a positive
economic return compared to those who do not return to finish a degree. Additionally,
this research examines whether their outcomes are impacted by the sector (either public,
private non-profit, or private for-profit) of the college or university at which they
complete their degree. Finally, this study examines whether individual characteristics
affect the likelihood that an individual who has stopped out of college will return to
complete a degree. Overall, I find that the economic return varies across racial/ethnic
background and that not all subgroups earn a positive return from finishing a degree, but
returns do not differ by sector. Finally, I find that many of the factors generally
associated with increased educational attainment do not appear to have a relationship
with the likelihood of finishing a degree.
The form and content of this abstract are approved. I recommend its publication.
Approved: Professor Paul Teske


IV
DEDICATION
For Sara, without whom this simply would not have been possible. And for Cora, without
whom this still would have been possible (and been possible quite a bit sooner), but
whose absence would have made life much less enjoyable.


V
ACKNOWLEDGEMENTS
This dissertation owes its existence to the guidance, support, and encouragement
of numerous individuals. Dr. Paul Teske guided this work and provided well-balanced
leadership, advice, and direction throughout the process as my advisor. Dr. Todd Ely also
provided thoughtful feedback and patient explanations of countless quantitative concepts
from the very inception of this research both in his faculty role and as the chair of the
dissertation committee. Drs. Kelly Hupfeld and Susan Clarke helped fine-tune the
research and pushed me to consider several additional angles that have resulted in a much
stronger overall research effort.
This work would also not have been possible without the support and advice of
colleagues at the Western Interstate Commission for Higher Education (WICHE), where I
have been employed throughout this process. WICHEs president, Dr. David
Longanecker, has established an organizational culture in which all staff are encouraged
to pursue additional education, while Dr. Demaree Michelau was extraordinarily
supportive throughout the project and an invaluable source of advice and guidance. Dr.
Brian Prescott has also contributed through encouragement and support, including in
helping to obtain the data necessary to complete the research.
Finally, this work owes a debt of gratitude to Dr. Peter deLeon. Though he may
not be aware, the initial meeting I had with him six years ago convinced me that pursuing
a Ph.D. was, all in all, a worthwhile way to spend ones time.


VI
TABLE OF CONTENTS
CHAPTER
I: INTRODUCTION.........................................................1
Significance of the Study in Practice.................................3
Outcomes at Public and Private Colleges and Universities..............8
Practical Implications Conclusion..................................11
Research Questions...................................................12
II: LITERATURE AM) HYPOTHESES..........................................14
The Economic Return to Degree Completion.............................15
Distinctions between Public and Private Organizations................35
Literature Review: Conclusion........................................42
Hypotheses...........................................................43
Contributions to the Field...........................................46
III: DATA AND MEASUREMENTS.............................................53
Data Source..........................................................53
Data: Conclusion.....................................................77
IV: METHODOLOGY........................................................78
Individual-Level Fixed Effects.......................................78
Calculating Rate of Return...........................................83
Event History Analysis...............................................88
Methodology: Conclusion..............................................91
V: RESULTS.............................................................94
The Wage Premium for Degree Completion...............................94


vii
Returns to Degree Completion.....................................104
Why Do Near Completers Return?...................................117
VI: DISCUSSION & CONCLUSIONS.......................................126
Economic Returns to Degree Completion............................127
Why Do Near Completers Return?...................................134
Conclusions......................................................137
REFERENCES.........................................................143
APPENDIX...........................................................157


Vlll
LIST OF TABLES
TABLE
1: College enrollment by age group and sector, 2011.............................11
2: Returns to degree completion.................................................19
3: Estimations of the sheepskin effect..........................................22
4: Research questions and hypotheses............................................45
5: College and state definitions for degree completion programs.................55
6: Descriptive statistics for those who attained near completer status...........58
7: Changes in descriptive data (unweighted) 1979 to 2012.......................60
8: Weighted descriptive data, 1979 and 2012.....................................60
9: Demographic data by educational attainment...................................64
10: Income by educational attainment 1992, 2002, & 2012.......................70
11: Income for near completers, by enrollment and graduation status.............70
12: Pre-enrollment income for returning near completers.........................71
13: Average tuition and fees by school management type, 1980-2012................73
14: Descriptive data for near completers who finish a degree, by sector..........74
15: Methodological approaches used to test hypotheses...........................92
16: Effects of education status on earnings.....................................97
17: Differential wage premiums by gender........................................100
18: Differential effects by racial/ethnic background............................101
19: Differential effects by 1979 poverty status.................................102
20: Income differences by sector of graduation..................................103
21: Approximated rates of return for baccalaureate degree completion............117


IX
22: Event history analysis of degree completion.......................................119
23: Event history analysis with gender interaction terms..............................121
24: Event history analysis with racial/ethnic interaction terms.......................122
25: Event history analysis with poverty interaction terms.............................123
26: Definitions of near completer....................................................157
27: Regressions with different definitions of near completer.........................158
28: Differential effects by gender....................................................160
29: Differential effects by race/ethnicity............................................161
30: Differential effects by familial poverty status...................................162
31: Income differences by sector of graduation........................................163
32: Event history analysis using different near completer definitions.................165
33: Event history analysis with gender interaction terms..............................166
34: Event history analysis with racial/ethnic interaction terms.......................167
35: Event history analysis sensitivity tests with poverty status interactions.........168
36: Categorical years since graduation................................................169
37: Continuous years since graduation.................................................170
38: Comparison of weighted and non-weighted regressions...............................172


X
LIST OF FIGURES
FIGURE
1: Near completers and near completers finishing degrees over time..............63
2: Near completers and near completers finishing degrees over time, accounting for
attrition.................................................................63
3: Survival estimates for near completers........................................66
4: Survival estimates for near completers by gender..............................67
5: Near completer survival rates by racial/ethnic group..........................68
6: Near completer survival rates by familial poverty status......................69
7: Plot of hypothetical rate of return for near completer finishing a degree.....87
8: Public tuition as a percentage of mean wages of near completers..............106
9: Non-profit tuition as a percentage of mean wages of near completers..........106
10: For-profit tuition as a percentage of mean wages of near completers.........106
11: Public tuition as a percentage of earnings of African-Americans/Hispanics...107
12: Return for baccalaureate degree completion - public tuition rates...........110
13: Return for baccalaureate degree completion - non-profit tuition rates.......Ill
14: Return for baccalaureate degree completion - public tuition rates...........112
15: Return for baccalaureate degree completion non-profit tuition rates.......113
16: Return for baccalaureate degree completion for-profit tuition rates.......113
17: African American/Hispanic return for baccalaureate degree completion public
tuition rates............................................................115
18: African American/Hispanic return for baccalaureate degree completion non-profit
tuition rates...................................................................115
19: African American/Hispanic return for baccalaureate degree completion for-profit
tuition rates............................................................116
20: African American/Hispanic rate of return with varied indirect costs.........171


XI


1
CHAPTERI
INTRODUCTION
Within higher education and workforce development policy circles, there is a
heavy focus on increasing the percentage of the United States population that has a
postsecondary degree or certificate. These efforts include a wide range of public sector
programs and policies designed to increase the rates at which individuals choose to attend
college and also boost the percentage of those who finish credentials once they enter.
There is strong evidence that increased degree attainment rates have societal benefits as
well as financial returns to the individuals completing degrees (see for example, Hout,
2012). Within a wide range of policy efforts to increase postsecondary credential
attainment, many state governments are including a special focus on those individuals
who earned significant college credits before leaving, working on the intuitive
assumption that they will be able to complete degrees and certificates more quickly and at
a lower cost (see for example Minnesota State Colleges and Universities, 2015;
Oklahoma State Regents for Higher Education, n.d.). Underlying this work is an
assumption that these individuals who finish degrees will earn substantial financial
benefits; however, little research has examined the question. Further, few, if any, studies
examine the factors that increase the likelihood that these near completers a term I
use to denote those individuals who finish significant college credits but do not earn a
degree will return to finish baccalaureate degrees.1
1 Although many policy efforts include a focus on encouraging those with prior college credit to return to
finish associates degrees or certificates, this study focuses on the economic return to baccalaureate degrees
only.


Although the general question about whether or not there is an economic benefit
to education may be settled, research on near completers is an unexplored avenue. While
2
labor economists have reached general agreement that earning a bachelors degree is
likely to produce an individual economic return, there has been no previous research
looking at the outcomes of near completers who finish degrees. Many of the data cited to
support these programs (and to encourage near completers to enroll) compare incomes for
any bachelors degree holder with non-degree holders, which is not the appropriate
comparison. The appropriate way to analyze income gains by near completers is to
compare their earnings after they graduate to near completers who had completed a
similar amount of postsecondary education. In addition to addressing this key question
for those returning to complete baccalaureate degrees, this study builds on public affairs
literature on public and private organizations by analyzing whether the economic benefits
earned by near completers who finish baccalaureate degrees are dependent on whether
they return to a public, private non-profit, or private for-profit college or university.
Finally, in this dissertation I explore the factors that affect the likelihood that a near
completer will return to finish his or her baccalaureate degree, which is of interest for
education and workforce development researchers as well as policymakers and
practitioners.
This study is organized as follows. After this brief introduction to the topic, the
importance of this research to the higher education and workforce development policy
community is discussed, noting several important gaps in empirical knowledge that will
be filled by these results. Following this introduction, I present formal research questions,
completing Chapter 1: Introduction. Chapter 2: Literature and Hypotheses consists of a


3
review of relevant literature, followed by hypotheses derived from the literature. Chapter
3: Data and Measurements focuses on describing the data and measures used for the
analysis. Chapter 4: Methodology presents the methodological approaches used to test
my hypotheses. Chapter 5: Results presents the results of the empirical analyses, and
Chapter 6: Discussion and Conclusions concludes the study with a discussion of the
results, their implications, and avenues for further study.
Significance of the Study in Practice
This research contributes to both the labor economics literature on human capital
and signaling, as well as public affairs literature on public and private organizations.
Additionally, this research informs current policy debates on degree completion
programs. In this section, I discuss the relevance to policymakers and practitioners by
providing some background on degree completion and near completers as well as the
programs being implemented to bring them back to college. The contribution that this
proposed study will make to the academic literature is discussed following the review of
literature in Chapter 2.
Increasing societal degree attainment. There is currently a strong push in
education and workforce development policy circles to increase the percentage of the
adult population in the United States that has a postsecondary degree or certificate. At the
societal level, arguments for this increased degree attainment generally hinge on
projections about future workforce demands and degree production rates, while
arguments aimed at individuals emphasize increased wages and better life outcomes.
The societal-level imperative for increasing degree attainment rates is based on
research showing that employer demand for individuals with college degrees will outstrip


4
degree production, resulting in a substantial degree gap by 2025 between the number of
jobs that will require postsecondary education and the number of individuals with
adequate qualifications (Camevale, Smith, & Strohl, 2010). This projected gap has driven
the setting of ambitious national and state goals for degree attainment. Lumina
Foundation, which is a large, private foundation focused on postsecondary education
outcomes, established a goal of having 60 percent of the adult population with a
postsecondary degree or certificate by 2025 to provide the necessary educated workforce
to meet future employment demands (Lumina Foundation, 2011). About 50 percent of the
adult population currently has a degree or certificate (Ewert, 2013; Lumina Foundation,
2013). The Obama administration established a similar goal of having the highest
percentage of 25-34 year olds with postsecondary credentials in the world by 2020 (U.S.
Department of Education, 2011). Currently, the U.S. ranks 16th (White House, n.d.).
Numerous states and even cities have also set aggressive goals for increasing degree
attainment (see for example 55,000 Degrees, 2014; HCM Strategists, 2014).
Increasing degree completion by adults and near completers. Efforts to
increase degree completion by adults generally thought of as being over 25 have
grown because it will not be possible to meet the goals cited above even with massive
(and highly improbable) improvements in the traditional education pipeline (those
individuals that go directly from high school to college and complete degrees) (National
Center for Higher Education Management Systems and Delta Project on College Costs,
2011). Out of the broader adult population, policymakers have naturally focused on the
22 percent of the working age population that falls into the some college, no degree 2
2 There is some debate about this figure because current data systems do not adequately track certificates,
which have a broad definition of any non-degree credential.


5
census category, meaning that they have finished some college credits without
completing a degree (U.S. Census Bureau, 2011). The logical conclusion is that these
individuals would have an easier time completing a degree than those starting from zero
credits. This broad census category includes all those who finished any college credits,
meaning that individuals in this category may have only finished one class, while others
may be within a few credits of a degree.
The strategy of targeting near completers is spreading. Many states are now
pursuing programs explicitly aimed at this population, but the assumptions behind these
efforts are untested. Chief among these is the assumption that near completers who finish
degrees receive an individual economic benefit (see for example Kentucky Council on
Postsecondary Education, 2008; Oklahoma State Regents for Higher Education, n.d.).
Messages based on this assumption tend to form the foundation of marketing and
outreach efforts that encourage near completers to return to postsecondary education to
finish a degree (see for example Adult College Completion Network, n.d.; Minnesota
State Colleges and Universities, n.d.). This conventional wisdom tends to be based on
bivariate models that compare income levels for those in the some college no degree
and bachelors degree categories from U.S. census data (see for example Baum, Ma, &
Payea, 2013).
This comparison is inadequate for several reasons. First, as noted earlier, this
category includes individuals who have earned any credits. Those who have earned
relatively few credits may have different characteristics than those who earned significant
credits and would likely pull down the average wages of this group. Second, given that a
substantial portion of the Bachelors degree category is likely to have completed degrees


6
on a more direct path than near completers, it is tenuous to assume that near completers
will realize a similar wage premium from finishing their degree. Third, these comparisons
do not usually examine important individual characteristics that may also affect the
earnings of near completers who finish degrees. A better comparison is to examine how
near completers who finish degrees fare economically compared to near completers who
do not finish degrees, while appropriately controlling for individual characteristics that
may also influence earnings.
Beyond uncertainty about the economic benefits for individuals who return to
finish degrees, there are several additional shortcomings in the research on which these
programs are based. There is little research on the credit distribution and demographic
details of the broader some college, no degree census category, meaning that it is
nearly impossible to tell how many are close to earning a degree (the near completers)
and how many earned only a few credits before leaving postsecondary education. In spite
of this uncertainty, many states are assuming that there are large enough numbers of
residents close to earning degrees to quickly raise attainment levels. Further, research has
not yet examined whether the types of degrees these near completers are likely to earn
will match future workforce needs to address the degree gap. Finally, some state
programs focus on the idea that near completers who finish degrees will increase tax
revenue through their higher salaries, which is another unproven assumption (Abdul-
Alim, 2011).3
3 Although it is beyond the scope of this study, the assumption that increases in individuals wages
automatically leads to increased tax revenue is flawed. If an individual earns higher wages, he or she will
pay higher taxes, but it could be that this individual took a job that some other person would have taken.
Overall taxes collected by a government will only increase if the aggregate wages in that economy increase.


7
One recent report examines those who have completed two or more years of
progress toward a degree within the last 20 years (Shapiro, Dundar, Yuan, Harrell, Wild,
& Ziskin, 2014). This research is informative as it compares this group, which Shapiro, et
al., term potential completers, with others who have completed far less postsecondary
education but still fall into the some college no degree category. While this study
shows that the potential completers make up about 11 percent of the some college, no
degree group, it does not track these individuals longitudinally to determine graduation
patterns and is only able to provide gender among demographic characteristics (Shapiro,
etal., 2014).
Policies and programs focused on near completers have also generally not
considered how such efforts impact different sub-populations. Attention elsewhere in
higher education policy circles focuses on reducing the gap in degree attainment between
racial/ethnic groups and providing additional supports to low-income students (Bailey &
Dynarski, 2011; Holzer & Dunlap, 2013). Without simple demographic data about near
completers, it is not possible to say whether these programs could help reduce
racial/ethnic and income disparities and degree attainment or whether such programs may
have different impacts on important subpopulations. Further, research has not examined
the characteristics of near completers that affect the likelihood that they will return to
finish a degree. Such information would help policymakers adjust programs, target
messages, and provide important supports to increase the number of near completers who
return to finish degrees. While this research does not fill all of the gaps facing
policymakers, it is an important step forward in providing key information about near
completers by examining the economic return earned by those near completers finishing


8
baccalaureate degrees as well as the factors that affect the likelihood that near completers
will return to finish such a degree. Although near completers who may return to complete
associates degrees are an important component of policies and programs in this area, this
population is beyond the scope of this study.
Outcomes at Public and Private Colleges and Universities
Within higher education research and policy circles, there has also been
substantial attention paid to the differences in outcomes between public and private
colleges and universities (Cellini & Chaudhary, 2014; Jacobs, 2013; Monks, 2000;
Schlesinger, 2010). Examining the outcomes for near completers who finish degrees at
different types of colleges and universities will also be an important contribution to the
policy conversation about degree attainment. Leaving aside, for the time being, more
nuanced characterizations about what makes an organization public, private, or
something in between, there are important discussions within higher education policy
circles about the outcomes of students who attend public, private non-profit, and private
for-profit colleges or universities.
As one example, a current debate focuses on the concept of undermatching,
where high achieving, low-income students tend to enroll in non-selective public or
private for-profit colleges or universities instead of attending selective, often private,
non-profit schools for which they qualify (Bastedo & Flaster, 2014; Hoxby & Avery,
2012; Hoxby & Turner, 2013). Interest here focuses both on the cost of attendance, which
can often be minimal at private, non-profit colleges and universities due to grants and
scholarships offered to low-income, high achieving students, and the graduation rates,


9
which some research suggests would be higher for these students at the more selective
schools (Hoxby & Avery, 2012; Hoxby & Turner, 2013).
In another example, the U.S. Senate Health, Education, Labor, and Pensions
Committee has held several hearings and issued multiple reports since 2010 about the
role of private, for-profit colleges and universities in educating students (U.S. Senate
Health, Education, Labor, and Pensions Committee, 2012; U.S. Government Printing
Office, 2010). The committee has focused on whether these private, for-profit colleges
and universities, which receive indirect federal funding through federal student aid
programs, provide adequate outcomes for students at a reasonable cost. One particular
thread of this work is how military veterans fare when entering for-profit colleges and
universities. These veteran students tend to be older than traditional students (having
served in the military for several years), and have relatively generous tuition benefits
through the Post 9/11 GI Bill that can make them attractive students for colleges and
universities that are highly dependent on tuition. Finally, the Obama Administration has
implemented regulations for colleges and universities that will limit the federal financial
aid dollars they can receive if their graduates fare poorly after earning a career-focused
degree (U.S. Department of Education, 2015). These regulations are aimed at reducing
the number of graduates from low-performing for-profit schools who pay substantial
tuition but are not able to find jobs that pay sufficient salaries to repay their debts (U.S.
Department of Education, 2015).
These examples focus on concern about differential outcomes between public and
non-public colleges and universities. This concern is reflected in the popular media as
well and tends to focus on the tuition prices for private non-profit and private for-profit


10
schools, which are usually higher than state-run public colleges and universities (College
Board, 2013; Jacobs, 2013; National Center for Education Statistics, 2013; Schlesinger,
2010). Additionally, policymakers have raised concerns that those who attend private for-
profit colleges and universities have lower graduation rates and higher student loan
default rates, indicating that they may be earning less than graduates from other sectors
(U.S. Senate Health, Education, Labor, and Pensions Committee, 2012). The
counterargument is that these outcomes are primarily due to selection bias and these
institutions tend to enroll low-income students, or those who are less well-prepared
academically and would have lower earnings no matter the sector from which they
graduated (Deming, Goldin, & Katz, 2011). Although some research has examined the
issue for first-time students, little, if any, attention has been paid to how public, non-
profit and for-profit colleges and universities may differ in how they serve returning
students and whether the outcomes by near completers may differ.
In addition to the statewide degree completion programs discussed above that try
to attract near completers to public schools, many private non-profit and private for-profit
colleges and universities are also engaged in similar programs (see for example Bellevue
University, 2013; University of Phoenix, 2015). There has been no research to show the
sectors in which near completers are most likely to enroll. Data on adult enrollment,
however, can shed some light on the question (assuming that most of these students fall
into the adult category, given that they have spent multiple years away from
postsecondary education). Those enrollment data show that a large number of adults are
enrolling in private schools (see Table 1). These data may not perfectly fit the population
of interest, but they do suggest further study is warranted into the outcomes of near


11
completers by sector. Based on these data, it appears that enrollment patterns may shift in
relation to age, with older individuals shying away from private, non-profit schools and
being more likely to attend either public or private for-profit ones. Although this is just an
approximation of enrollment rates for near completers, research that can identify whether
the outcomes for those students vary by institutional type would contribute significantly
to the ongoing policy discussions surrounding public and private schools.
Table 1: College enrollment by age group and sector, 2011
Age Group Public College or University Private, Non-profit College or University Private For-profit College or University
Under 24 59.7% 28.2% 12.0%
25-29 69.0% 18.7% 12.4%
30-39 64.5% 16.6% 18.9%
40 and over 64.0% 16.2% 19.8%
Source: National Student Clearinghouse, 2012a, 2012b
Finally, although there has been some limited market research that examines the
types of marketing and outreach messages that appear to resonate with near completers,
there has been little formal research into the characteristics that are associated with
returning to finish a degree (see for example EducationDynamics, 2010; Maguire and
Associates, 2010; Minnesota State Colleges and Universities, 2013). Better understanding
the individual characteristics that may lead near completers to return to postsecondary
education, as well as those that may prevent them from returning, could help
policymakers better design programs to serve this population.
Practical Implications Conclusion
The research that follows cannot fill all of these gaps in the policy environment
surrounding near completers. However, by examining the economic returns of near


12
completers, the individual characteristics that may affect those returns, the educational
outcomes of near completers by sector, and the factors that affect the likelihood of a near
completer finishing a degree, this study contributes greatly to ongoing efforts to increase
the attainment rates of this group.
Research Questions
Thus, while there is a strong policy argument behind the need to increase degree
completion by near completers, there is a need for rigorous research to substantiate many
of the assumptions underlying such programs. State governments, as well as colleges and
universities are committing millions of dollars to programs aimed at serving these
students, while large numbers of college stop-outs are spending tens of thousands of
dollars and sacrificing substantial time to finish degrees all based on these untested
assumptions. Based on the gaps discussed above, this dissertation will examine the
following questions:
RQ1: Do near completers who return to finish a baccalaureate degree earn a
positive economic return compared to near completers who do not return?
RQ2: Does the sector of the college or university from which a near completer
graduates affect his/her economic return?
RQ3: How do the characteristics of individual near completers affect the
likelihood that they will return to a college or university and complete a baccalaureate
degree?
Answering these research questions will make contributions to higher education
and workforce development practitioners by evaluating a key claim of a major public
sector effort and by providing more information about the types of students that current


13
efforts are successfully reaching. As detailed more fully in the following section, this
study will also contribute to academic research in labor economics and public
administration. Hypotheses for each research question are derived from the review of
literature and are presented in Chapter 2.


14
CHAPTER II
LITERATURE AND HYPOTHESES
The literature that informs these research questions comes from different fields.
The basic question of whether a near completer who returns to finish a baccalaureate
degree is likely to earn a positive economic return relies significantly on human capital
and signaling theories from the field of labor economics. The second research question
draws on public management literature that focuses on the relationship between public
and private management of an organization and outcomes of public programs, as well as
literature on outcomes by sector. This leads to a review of literature on the outcomes of
students at public, non-profit, and for-profit colleges and universities.4 The final research
question draws on the limited research about the factors that are associated with
individuals (particularly adults and non-traditional students) pursuing additional
education and training.
The literature review is structured accordingly, beginning with an overview of
research on returns to education before focusing more specifically on returns to adult
education and the limited studies available on the returns to near completers. Following
this, the review examines potential sources of bias in research on returns to education and
methodological approaches to addressing them. This discussion includes an examination
of the factors that affect decisions to pursue additional education and training. The review
concludes with an examination of the research on the outcomes produced by public and
private organizations generally, and more specifically, the existence of research on
4 Although traditionally in higher education literature these sectors are referred to as public, private, and
for-profit (or proprietary), to be consistent with public administration literature, non-profit will be used in
place of private throughout.


15
differential outcomes between public, non-profit, and for-profit colleges and universities.
As noted above, this study focuses exclusively on near completers who return to
complete baccalaureate degrees. However, the literature review draws on studies on all
types of postsecondary credentials due to the limited extant research on the population of
interest.
The Economic Return to Degree Completion
Labor market economics as a field has developed an immense body of literature
related to the individual financial benefits of education. This section of the literature
review begins with a discussion of the foundations of human capital theory and the
empirical model that forms the basis for estimates of the impact of schooling on wages,
followed by a discussion of the theoretical origins of signaling theory and the so-called
sheepskin effect. The review then turns to research specific to the economic returns to
college completion, followed by a review of the limited research on degree completion by
near completers. This section finishes with a discussion of methodological refinements in
the literature that have helped to reduce potential sources of bias in estimating the
economic benefit of additional education.
Human capital theory. Literature on human capital theory forms the foundation
for answering the question of whether near completers who return to finish a degree will
receive an economic benefit. Mincer (1958) and Becker (1993 [1964]) are generally
viewed as providing the key beginning points for human capital theory even though both
authors build on a number of predecessors in economics literature (see for example
Moore 1970 [1911]; Staehle, 1943). Mincers eponymous equation has become the
standard method of estimating economic returns on schooling. In the Mincer equation


16
earnings rise with age and experience, but then begin to decline at older ages (Mincer,
1958). The amount of schooling and training individuals receive also affects their
beginning salaries and impacts the slope of lifetime earnings, meaning that those with
more schooling are likely to see their income continue to increase relative to those with
lower levels of education (Mincer, 1958). The standard Mincer equation is as follows:
In Y = In Yo + fas + + ft3^
Where Y is an individuals earnings, Y0 is the earnings of an individual with no schooling
or experience, 5 represents years of schooling, and e represents work experience (Mincer,
1974). The quadratic experience term accounts for the changes in the earnings function as
experience grows. Mincer also came to include a vector of individual characteristics that
also affect earnings including, racial/ethnic background, sex, family status, and
characteristics of the city in which the individual lives (Mincer, 1958, 1974).
While this equation addresses the relationship between earnings and a variety of
factors, it relies on a series of assumptions in order to provide an actual rate of return
rather than just the wage premium for an additional year of schooling. The equation
assumes that there are no direct costs associated with schooling, that an individual will
work the same length of time independent of his or her schooling level, that education
and experience can be estimated separately (i.e. they do not have a combined effect on
earnings), and that there are no income taxes (Ashenfelter, 1978; Bjorklund &
Kjellstrom, 2000; Card, 1999; Heckman, Lochner, & Todd, 2008; Heckman, Lochner, &
Todd, 2006). The Mincer equation also focuses on years of schooling and assumes a
linear relationship between years of schooling and the log of earnings, without deviations
from that linear path that can be evident for completing certain credentials like high


17
school diplomas and college degrees (i.e. the 12th and 16th years of schooling) (Card,
1999; Heckman, Lochner, & Todd, 2008). Rates of return estimated using a Mincer
equation generally do not incorporate discount rates because they apply equally to
earnings for both the treatment and control levels of education, so the discount can be
dropped from consideration. Additionally, there are several difficult-to-measure variables
omitted from the equation, such as innate ability, ambition, health, and parental
education, that can bias estimates, although the amount of bias may be quite small
(Becker, 1993). Discussion of ability bias and selection bias in studying the returns on
education continues to the present day and is presented in greater detail below.
Mincers equation, with its assumptions, is a special case of a more general
equation proposed by Becker (1993) (Heckman, Lochner, & Todd, 2006). Beckers
equation incorporates costs, and because costs are borne at the time education is received,
he also incorporates a discount rate. Beckers equation, adapted from Heckman, Lochner,
and Todd (2006) is as follows:
^t=r(i+ r)t-2,t=o (1 + r)t
yT0 Ypt + Ct
Lt=o (i + ry
In this equation (with the explanation adapted for the purposes of this study), Yj
represents the earnings for an individual with a near completer who has finished a
baccalaureate degree at time t, while Y0 represents the earnings for a near completer at
time t; r represents a discount rate, while t represents the number of years an individual is
enrolled in college. The first numerator term sums the total earnings starting at t=x, which
means it only counts earnings after schooling is complete. The second numerator term
(and the identical denominator) sum the total earnings from t=0. The practical effect of


18
this is that an enrolled individual earns nothing while pursuing a degree (from t=0 to t=x),
which is Beckers method for accounting for foregone wages. The direct costs for
attending college are included in the term C) where C represents direct costs per year to
complete a degree. Once the individual has finished his or her degree this term equals
zero.
While less parsimonious than Mincers equation, Beckers treatment captures the
importance of including direct costs. Under this formulation, the economic return for a
near completer finishing a degree is positive if his or her wages (appropriately
discounted), subtracting costs of attendance, are greater than the wages of a near
completer who does not return to school. Although this equation does not account for the
possibility that near completers who are working on degrees may also be working at the
same time, it forms a more complete starting point for determining whether there is a
positive return for degree completion. The models proposed in Chapter 4 combine
aspects of both approaches to address the key research question about returns to degree
completion.
Wage premiums and postsecondary education. There is strong agreement within
the literature that additional years of schooling are associated with increased earnings
using a variety of methodologies, data sources, and approaches (Ashenfelter & Rouse,
1998; Ashenfelter, Harmon, and Oosterbeek, 1999; Hausman & Taylor, 1981; Hout,
2012; Ashenfelter & Krueger, 1994). Given this studys focus on degree completion, the
literature review will not focus on years of schooling in detail, but rather the wage
premium for degree completion.


19
The empirical evidence that individuals receive a wage premium for completing a
college degree is also relatively strong (Cellini & Chaudhary, 2014; Grubb, 1997; Jepsen,
Troske, & Coomes, 2014; Kane & Rouse, 1995; Marcotte, Bailey, Borkoski, & Kienzl,
2005). The estimates of the premium vary somewhat across studies as researchers use
different comparison groups, disaggregate data differently, and examine different
subgroups of interest. An overview of the some of the estimated returns is included in
Table 2.
Table 2: Returns to degree completion
Degree Grubb (1997) Kane & Rouse (1995) Marcotte, et al. (2005)
Level Male Female Male Female Male Female
Assoc.Degree 18.1% 22.8% 26.6% 20.7% 17.1% 40.4%
Bach. Degree 54.8% 53.4% 52.5% 39.2% 45.8% 91.9%
Above Bach. See Below 95.2% 53.1% N/A N/A
Masters 64.9% 77.9% N/A N/A N/A N/A
Prof. Degree 174.6% 153.7% N/A N/A N/A N/A
Ph.D 122.6% 141.3% N/A N/A N/A N/A
Note: Grubb includes measures for Masters degree, professional degree, and Ph.D, separately.
Kane and Rouse group these together. Marcotte, et at, include only associates degrees and
bachelors degrees.
Wage premiums are not the only economic benefit that those who finish college
degrees receive. Degree holders also have lower unemployment rates and greater
buffering through economic downturns (Gangl, 2006; Hout, 2012; Hout, Levanon, &
Cumberworth, 2011; Jepsen, Troske, & Coomes, 2014). These additional effects could
form a meaningful portion of the overall monetary benefit that an individual receives for
returning to complete a postsecondary degree, depending on how wages are calculated
and what controls are used for employment. One estimate suggests that 2/3 of the return
to education is due to higher wages, while 1/3 is due to more hours worked (Card, 1999).
This suggests that using annual wages as an outcome variable may be prudent to capture
the full effect of education on earnings.


20
Heterogeneous effects of degree completion. The benefits of degree completion
may not accrue to members of different sub-groups equally. The effects of education may
be heterogeneous across certain important characteristics such as racial/ethnic
background, socio-economic status, and gender. Individuals of lower socio-economic
status may actually benefit more from postsecondary education than those from more
advantaged backgrounds, as research shows that those who are least likely to graduate
from college realize a 30 percent wage gain, while those who are most likely to graduate
realize a 10 percent wage gain (Brand and Xie, 2010). These findings counter the idea of
positive selection bias, which suggests that those individuals who are most likely to earn
higher wages anyway are also most likely to attend college (see for example, Carneiro,
Hansen, and Heckman, 2003). Other research finds different outcomes by gender,
racial/ethnic background, number of children, and other factors (Bailey, Kienzl, &
Marcotte, 2004; Bitzan, 2009; Henderson, Polacheck, & Wang, 2011; Jaeger & Page,
1996). This suggests that the individual characteristics of near completers may affect
their economic return.
Signaling theory. Signaling theory holds that an individuals education level
serves as a signal to potential employers about his or her abilities and characteristics,
such as potential productivity, work ethic, persistence, and more (Arrow, 1973; Weiss
1995). Individuals pursue a level of schooling commensurate with these abilities, thus
indicating their relative value in the labor market. Higher incomes, then, may not be due
to the additional years of schooling, but to the signals sent by an individuals credentials.
Under signaling theory, employers use education as a screen to hire the most desirable
workers those who are less likely to be unhealthy (and thus miss work), less likely to


21
leave the job, and more likely to avoid habits such as excessive drinking and drug use -
provided these characteristics are correlated with the amount of schooling an individual
possesses (Weiss, 1995). Although related to human capital theory, this presents a
different argument in that individuals obtain a certain amount of education to send certain
signals to employers, rather than improving their potential productivity through learning.
Signaling theory is particularly manifest in research examining the sheepskin
effect. This is a phenomenon where a particular educational credential (conferred through
a diploma, hence sheepskin) is found to have an economic value over and above what
would be expected from an equivalent amount of schooling. As an example, someone
who completes 16 years of school but no college degree would be predicted to earn less
than someone with the same amount of schooling but has received a bachelors degree.
Research has confirmed that the sheepskin effect exists (empirical results are discussed in
greater detail below) supporting signaling theory and accounting for a portion of the
wage premium received by college graduates, although this effect does not appear to
apply to all groups equally. A related view is that individuals considering how much
education to obtain may make this decision based on their own views of their earning
potential so that they can accurately signal this potential without incurring unnecessary
extra costs in tuition or foregone earnings (Carneiro, Hansen, and Heckman, 2003).
Based on this view, workers who feel that their jobs are not appropriately matched to
their abilities and characteristics could try to finish a degree in order to send a different
signal to their potential employers.
Empirical evidence for the sheepskin effect. While much of the research cited
previously about the wage premium for college completion does not specifically test for


22
the sheepskin effect, studies that do explicitly examine this phenomenon provide further
evidence that individuals who earn college degrees receive a wage premium. Controlling
for years of schooling, several studies show that college diplomas generate a wage
premium for individuals (Bitzan, 2009; Hungerford & Solon, 1987; Jaeger & Page, 1996;
Jepsen, Troske, & Croome, 2014; Park, 1999). Other results for estimations of the
Table 3: Estimations of the sheepskin effect______________________________________________
Author(s), year Findings
Bailey, Kienzl, & Marcotte, 2004 Insignificant sheepskin effects for males at all degree levels; 28% wage premium due to occupational associates degree completion for women, insignificant for other degrees.
Belman & Heywood, 1991 21% wage premium for degree completion by racial/ethnic minority males, compared to 10% premium for white males. 25% wage premium for degree completion by racial/ethnic minority females, compared to statistically insignificant premium for white females. For lower-level signals (i.e. grade school and high school) racial/ethnic minorities had lower estimated sheepskin effects than white individuals.
Bitzan, 2009 20% wage premium for bachelors degree completion by white males; 14% wage premium for bachelors degree completion by black males. Black males receive higher premiums for graduate degrees than white males.
Hungerford & Solon, 1987 9% wage premium for bachelors degree completion.
Jaeger & Page, 1996 Earlier estimates of sheepskin effect are biased because data did not include both degrees and years of schooling, but relied on imputing degrees. 25% of wage premium from bachelors degree is due to sheepskin effect. No statistically significant differences between effects on different sub-groups. 21% wage premium for white males who earn associates degrees, statistically insignificant for black males. 27-36% wage premium for white females who earn associates degrees, statistically insignificant for black females.
Jepsen, Troske, & Croome, 2014 74% of wage premium for associates degrees for males is due to diploma, not credits; 65% of wage premium for female associates degree earners is due to the diploma.
Kane & Rouse, 1995 Small evidence of sheepskin effect for bachelors degree completion by males and associates degree completion by females, but statistically insignificant for other combinations.
Park, 1999 21% wage premium for bachelors degrees compared to those with 16 years of schooling but no degree; 11% wage premium for associates degree earners compared to those with 14 years of schooling but no degree.


23
sheepskin effect are mixed, with only some subgroups receiving a benefit or findings of
no effect (Bailey, Kienzl, & Marcotte, 2004; Kane & Rouse, 1995). A comparison of
results is presented in Table 3.
There is general agreement that the sheepskin effect is real and does provide a
wage premium to some of those who complete degrees, but it may not be present for all
subgroups. Some of the differences in the estimated size of the sheepskin effect are likely
due to differences in selected data and methodological approaches. Research only
focused on identifying non-linearities in the overall return to schooling for particular
years in which individuals earn diplomas or credentials (for example, the 12th and 16th
years of schooling) may be overestimating the sheepskin effect (Skalli, 2007). Estimates
for the size of the sheepskin effect that incorporate both years of schooling and actual
degrees earned are lower than those that only use years of schooling (Bitzan, 2009;
Jaeger & Page, 1996). Still, even with the differences in estimates, the existence of the
sheepskin effect could suggest that near completers who finish a degree are likely to
receive a wage boost even if they only need a small number of credits to finish a degree.
Heterogeneous sheepskin effects. As can be seen in Table 3, some research on
the sheepskin effect also finds heterogeneous returns for different population sub-groups.
Belman and Heywood (1991) find that women and racial/ethnic minorities have greater
sheepskin effects than white males. Jaeger and Page (1996), however, suggest that this
finding could be biased and that with proper measurement of degree completion, there is
no evidence that there are differential returns to diplomas. Kane and Rouse (1995) find
only a small sheepskin effect for men completing a baccalaureate degree and women
completing an associates degree, but an insignificant result for other combinations. They


24
further suggest that the majority of womens sheepskin effect for associates degrees is
due to the earning potential of nursing degrees (Kane & Rouse, 1995). For bachelors and
associates degrees, Bitzan (2009) finds that white males receive larger sheepskin effects
than black males. But in examining degrees above the bachelors level, he finds that
black males receive a larger sheepskin effect (Bitzan, 2009). While this research is not
focused specifically on near completers, it does suggest that estimates of wage premiums
due to diplomas may differ by gender and racial/ethnic background.
Wage premiums for degree completion by near completers. There is limited
literature that examines the economic benefit of near completers who finish degrees.
There are some case studies examining the outcomes of students who finish degrees
through a college or university that offers a degree completion program for stopouts. The
population under consideration bears some similarities to near completers studied here,
but few studies specify the population beyond examining college stopouts. These studies
generally rely on post-completion surveys of program completers. Mishler (1983)
surveys graduates who completed a bachelors degree in Wisconsin after stopping out
and finds large numbers reported job-related improvements due to earning their degrees.
Green, Ballard, and Kern (2007) use a similar approach to survey graduates who returned
to complete degrees and again find evidence for positive career outcomes based on
degree completion. These findings are echoed by a series of methodologically similar
evaluations (Culver, 1993; Harris, 2003; Hoyt& Allred, 2008; McKinney, 1991). These
evaluative efforts are suggestive but do not employ relevant comparison groups and thus
differ significantly from the study carried out here.


25
Wage premiums for adult education and training. Another relevant thread in
literature on wage premiums for education compares what adults who pursue education
later in life might earn compared to those who complete their education and training
earlier, such as those students who attend and complete college directly after high school.
Elman and ORand (2004) find that the wage benefits for adults who earn credentials are
smaller (and in some cases non-existent) compared to those who earn degrees one a more
traditional pathway.5 While their work includes those adults who already have some
college credit, their inquiry focuses on the timing of educational attainment and does not
focus specifically on near completers. Additionally, they limit their estimation to hourly
wages as an income measure, which could mask some benefit as research suggests one
pathway through which education affects earnings is by increasing the hours an
individual works (Blanden, Buscha, Sturgis, & Urwin, 2012; Card, 1999). Even with
these caveats, this research shows that comparing near completers who return to
traditional students may not be appropriate and that there are differences in returns to
postsecondary credentials depending on when they are completed.
Light (1995) examines the returns for those who stop out of education and return
later compared to those who complete a similar amount of schooling but on a direct
pathway. Those individuals who return to school after a period in the workforce
experience wage gains, but they are less than those who complete their schooling on a
direct path (Light, 1995). This research does not focus on the level of education obtained
when the individual stopped out. Leigh and Gill (1997) follow a similar path in
5 Although terminology is shifting within the policy community to reflect the normalcy of adult
enrollment in postsecondary education, adults are frequently referred to as non-traditional students, while
those who pursue degrees directly out of high school are referred to as traditional students.


26
comparing adult community college graduates with traditional-aged graduates and find
adults actually receive a slightly greater benefit.
There is also research from abroad that examines outcomes for adults pursuing
any type of education. Research from Sweden finds that individuals who complete
degrees as adults have 18 percent higher employment rates and 12 percent higher
earnings (Hallsten, 2012). Other European evidence also suggests that adults who pursue
education may end up with better employment prospects (Kilpi-Jakonen, de Vilhena,
Kosyakova, Stenberg, & Blossfeld, 2012). These comparisons examine adults who
pursue additional education in comparison with adults who do not, which is a more
relevant approach for the current study (Hallsten, 2012).
There has been significant research on wage benefits received by adults who
finish additional education, but little of it focuses on near completers and none of it
focuses on whether their decision to return to finish a degree is likely to result in a
positive economic return. Although adults are not directly comparable to near completers,
the findings above do suggest that near completers who finish degrees may receive an
economic benefit. Without estimations of foregone wages and direct costs and without
comparisons to other near completers who do not finish degrees, it is not possible to
conclude from the literature that they are likely to receive a positive economic return.
Heterogeneous effects for adult learning. As noted above, researchers have
identified heterogeneous effects of education on different sub-groups. This is an
important factor in examining wage premiums for adults as well. Brand and Xie (2010),
as noted above, find strong evidence that individuals with lower socio-economic status
benefit more from postsecondary education than those from advantaged backgrounds.


27
Elman and ORand (2004), also test for differential returns for adults who complete
degrees. They find a statistically significant wage premium for women who complete
bachelors degrees, but none for men. Other researchers have also found evidence of
differential returns by gender for adults participating in education and training programs
(Blanden, et al., 2012; Hallsten, 2012). Blanden, et al. (2012) find that women who
complete credentials as adults receive a 10 percent wage premium while similar men
receive no premium. Similar to sheepskin effects, this may be due to different
credentialing requirements in fields in which men and women tend to work (Blanden, et
al., 2012). These findings, coupled with research on heterogeneous sheepskin effects by
gender and racial/ethnic background, suggest that economic returns for near completers
who finish degrees may differ by sub-group membership.
Sources of bias in estimates of the return to education. Human capital and
signaling literature also recognize key methodological considerations in estimating the
wage premium associated with additional education. As noted above, ability bias is a
crucial factor in determining whether increases in earnings are due to an individuals own
characteristics or the schooling or training received (Becker, 1993). Highly able
individuals may be more likely to pursue additional schooling as well as earn high wages
regardless of schooling so estimates that do not control for ability will be upwardly
biased (Becker, 1993; Card, 1999; Griliches, 1977). Many researchers believe, with some
empirical support, that although ability bias exists and leads to overestimations of the
return to education, the amount of bias is relatively small (Becker, 1993; Card, 1999;
Griliches, 1977). This section examines selection bias in general research on the returns


28
to education before turning specifically to issues surrounding this potential bias in adult
education and training programs.
Efforts to mitigate selection bias. Extensive theoretical and empirical work has
sought to eliminate ability bias as a concern in estimates of the return to education. The
original approach was to include control variables for measures of individual ability in an
ordinary least squares (OLS) estimation. Including standardized test scores from high
school as a control variable has been one approach, although this may not be fully
effective (Card, 1999). Additional controls for family background, high school
achievement, and demographic characteristics are also regularly used. However, concern
that these variables may not fully capture innate ability and motivation, thus not
eliminating ability bias, has led researchers to employ a range of other methodological
approaches to develop more accurate estimates for the effect of schooling on earnings.
Although it is not feasible to use experimental controls to assign individuals randomly to
higher or lower education groups to test for ability bias, researchers have tried to
approximate such conditions.
One such approach has been to study the wages of identical twins with different
levels of education to attempt to quantify the size of ability bias in OLS-based
estimations. This rests on the assumption that identical twins have the same innate ability
and family background, so if they have different levels of education, any differences in
their earnings would be attributable to schooling. Behrman and Rosenzweig (1999),
following this approach, estimate that ability bias leads to an overestimation of the impact
of additional schooling on earnings by 12 percent. Card (1999) carries out a meta-
analysis of samples using twins and finds a roughly similar bias of about 10 percent.


29
However, two other methodologies that aim to control for ability bias -
instrumental variables and fixed effects models have consistently resulted in larger
estimates of the economic benefit of education compared to OLS estimates that attempt
to control for bias through observed variables.6 Hausman and Taylor (1981) find a seven
percent wage premium for each additional year of schooling using OLS compared to a
nine percent wage premium using an instrumental variable. Ashenfelter, Harmon, and
Oosterbeek (1999) compare different methodologies and find that instrumental variables
and fixed effects lead to near-identical results. Their meta-analysis suggests that the wage
premium of an additional year of schooling using instrumental variables or individual-
level fixed effects is about nine percent per year of schooling, compared to seven percent
with OLS-based approaches using control variables to account for bias. Other research
supports these findings, concluding that using individual-level fixed effects models
produces estimates that are between 40 and 50 percent larger than other approaches
(Ashenfelter & Rouse, 1998; Ashenfelter & Krueger, 1994).
These findings have generated divergent explanations. Some suggest that this
shows evidence of negative selection bias, which holds that those who are least likely to
complete additional schooling actually benefit the most. Brand and Xie (2010) conclude
research must examine heterogeneous returns to education namely that different sub-
groups (particularly low-income individuals and racial/ethnic minorities that may be less
likely to complete higher levels of education) receive different economic benefits from
6 Both of these approaches attempt to mitigate bias due to omitted variables and approximate experimental
conditions. Fixed effects approaches, which are used in this study, are explained in greater detail in the
chapter on methodology. Instrumental variables try to correct for situations where independent variables
are correlated with the error term by identifying an instrument that is correlated with the independent
variable of interest but not the error term. For additional information, see Wooldridge (2006) and Angrist
and Pischke (2009). For the purposes here, the key fact is that these approaches can, in theory, greatly
reduce concerns that innate ability or other omitted variables are leading to bias in estimates of the
economic return to education.


30
additional learning and credentials. Others argue that OLS approaches using control
variables without instrumental variables or fixed effects may underestimate the wage
premiums for education due to measurement error that can negatively bias coefficients
(Deaton, 2010; Hout, 2012). An alternative explanation focuses on instrumental variables
and suggests that there could be publication and reporting bias as researchers are most
likely to select instruments that produce statistically and substantively significant results
(Ashenfelter, Harmon, and Oosterbeek 1999). With appropriate data sources, however,
individual-level fixed effects approaches control for all time-invariant characteristics of
an individual (Wooldridge, 2006). Thus, estimations based on fixed-effects models must
assume that an individuals innate ability does not change over time, but should this hold
true, they will properly account for ability bias. The chapter on methodology discusses
these characteristics in greater detail.
Selection bias and near completers. One important question in examining
whether near completers who finish degrees receive a wage premium is whether those
who finish are substantively different from those near completers who do not return. It is
possible that these differences, whether in work ethic, demographics, ability, or other
characteristics, may affect income in ways that could mistakenly be attributed to the
effect of degree completion. Again, with limited extant research on near completers,
examining the available research on adult learners can be a helpful starting point.
Motivation for pursuing additional education is one consideration. If adults are
pursuing additional education for non-economic means, it could be a factor if no wage
premiums are evident. Alternatively, it could be that those who are certain they would
receive workplace benefits if they complete a degree are much more likely to return and


31
finish their schooling. Houle (1961) conducted the foundational study on the motivation
of adult students and identified three major types of adults pursuing education: goal-
oriented individuals, activity-oriented individuals, and learning-oriented individuals.
Other research has identified professional advancement as well as pursuing learning for
its own sake and enjoying participation in group activities as reasons that adults pursue
additional education (Morstain & Smart, 1974). Further work has confirmed and
augmented Houles original study and concludes that adults pursuing education consist of
those who are economically motivated as well as those who are motivated by the
enjoyment of learning or desire to participate in group activities (Boshier & Collins,
1985). It may be difficult to ascertain adults motivation for pursuing additional
education and linking that to income data with available datasets. However, this could be
an important confounding factor in an analysis of near completers. There could be a
relationship between the motivation for pursuing education and resulting income that is
unrelated to the effect of additional schooling on income. For example, an individual who
becomes more ambitious later in life may decide to finish his or her degree and later
earns higher wages due to that increased ambition rather than the degree he or she
finished. It may be difficult, however, to separate the effect of education from changes in
ambition.
Other research has sought to identify some of the observable factors associated
with adults pursuit of additional education. Elman and ORand (2004) identify family
background and socio-economic status as key factors in adults decisions to pursue
additional education. Additionally, they find that having children at an early age, entering
the military, and belonging to certain racial/ethnic groups are also positively associated


32
with pursuing additional education (Elman and ORand, 2004, 2002). An alternative
explanation of these latter factors could be that individuals who had early children or
entered the military were more likely to leave education short of their desired attainment.
Age, marital status, family size, family income, and tuition levels are also important
factors in predicting adult enrollment (Light, 1996). While these factors appear to be
important for adult enrollment, further research is necessary to determine whether they
are applicable to near completers, as well.
The Ashenfelter dip. In considering the current study, one source of potential bias
is the income level of near completers prior to reenrollment. A large body of research
suggests that for both education and training programs, lower pre-treatment incomes are a
significant factor in predicting participation. These findings come from studies examining
both postsecondary education and workforce training programs (Ashenfelter, 1978;
Cellini & Chaudhary, 2014; Heckman & Hotz, 1989; Jepsen, Troske, & Coomes, 2014).
Ashenfelter (1978) originally identified the phenomenon that now bears his name. The
Ashenfelter dip refers to the fact that income for adults who enter workforce training
programs tends to decline immediately prior to the treatment. Following completion of
the training program, as an individuals income tends to naturally recover, estimations of
its impact may be overstated as this natural recovery is interpreted as a treatment effect
(Ashenfelter, 1978). This phenomenon may also be evident across a population during
periods of economic recession, in which adults may be more likely to seek out training
and education programs. Improved individual incomes following recessions are likely
due in part to natural economic recovery in addition to any benefits from the training and
education programs (Ashenfelter, 1978).


33
Others have found evidence of the Ashenfelter dip in training and education
programs. Heckman and Smith (1999) find a pre-training decrease in income for some
that participate in training programs. Marcus (1986) finds that adults whose earnings
decline below values predicted from relatively straightforward regression models are
more likely to return to school. Blanden, et al., (2012), using data from England,
conclude that credentials earned later in life (including the equivalent of U.S. high school
degrees, vocational certificates, and college degrees) have economic benefits for women,
but find that the effect for men disappears when measures are taken to control for trends
in pre-treatment income. Their work is slightly different in that they identify a dip
followed by the beginnings of an upward income trend for adults participating in
education and training programs for men, but not for women. Cellini and Chaudhary
(2014) and Jespsen, Troske, and Coomes (2014) also find evidence of an Ashenfelter dip
for adults who pursue community college degrees. Additional research has concluded that
pre-treatment income is an important determining factor in whether adults enroll in
further education, with lower-income adults more likely to enroll (Jepsen &
Montgomery, 2012; Light, 1996). Thus, it will be imperative to account for the income of
near completers prior to their return to postsecondary education. If there is an Ashenfelter
Dip, it would potentially bias estimates of the wage benefit for degree completion.
Foregone wages and indirect costs. As described above, accounting for foregone
wages is a central part of determining whether an individual receives a positive economic
return to degree completion. However, foregone wages could also be a source of potential
bias in the estimates of the wage premium for degree completion. If returning students
forego substantial wages while enrolled, their incomes would recover naturally when they


34
start working more after completing a degree, which could upwardly bias estimations.
Again, with no literature available specific to near completers, I turn to research on adult
learners instead. The literature on whether older students forego significant wages is
somewhat mixed. Jespsen and Montgomery (2012), who found that lower income
individuals are more likely to seek out additional education and training, argue that this
shows that opportunity costs are a significant barrier for adults to enter postsecondary
education. They suggest that those with higher incomes would pay higher opportunity
costs to complete additional education (Jepsen & Montgomery, 2012). However, they
present little data to conclude that when adults reenroll in postsecondary education they
forego earnings. An alternative explanation could be that those adults who are already
faring relatively well economically see less need to pursue additional education. Blanden,
et al. (2012) suggest that evidence of depressed wages prior to completing adult
education and training programs could be due to foregone wages or simply that
productivity is lower for these adults. Employment status prior to education has also
proved to be an important factor in assessing benefits of adult education in multiple
European countries, and studies there have shown some evidence that adults may forego
wages to pursue additional education (Kilpi-Jakonen, et al., 2012).
Thus, while initial salary may be a key determinant of decisions by adults to enter
postsecondary education, the literature suggests that adults forego wages to pursue
additional education and training. However, there are few studies that examine adult
college-going in detail, and it could be that programs are more flexible and allow these
students to keep working full-time while enrolled. Additionally, data show that adults
tend to opt for part-time enrollment and other flexible options offered by higher


35
education institutions (National Center for Education Statistics, 2014). Still, it will be
important to control for foregone wages so that any natural recovery following school
does not appear to be a benefit of degree completion. Additionally, estimations of these
foregone wages are necessary to account for the full cost of returning to finish a degree.
Distinctions between Public and Private Organizations
A central question in public administration and public management research has
been whether an organizations publicness affects the outcomes it produces (Bozeman,
1987; Walmsley & Zald, 1973). The roots of the focus in public administration on the
differences between public and private management are evident in the work of many of
the foundational writers in the field. This section of the literature review approaches the
question of public and private organizations by first examining the criteria used by the
field to delineate the two types of organizations, then illustrating how an operational
definition of public and private management has been used in research on higher
education, before concluding with a discussion of the literature on inputs and outcomes at
public and private colleges and universities.
Foundations of public vs. private debates. Perhaps starting with Taylors (1967
[1912]) principles of scientific management, which were originally developed for private
firms, the tension between public and private organizations is evident in public
administration. That Taylors work came to be applied to distinctly public government
agencies suggests that early in the field, there was a blurred distinction between the two
sectors. Weber (1970 [1922]), in his classic treatise on bureaucracy, argues that his
conclusions apply equally to the government and private sectors. Simon (1957) suggests
blurred lines between public and private organizations, arguing that assumed distinctions


36
may not apply within actual organizations. Dahl and Lindblom (2000 [1953]) proposed a
continuum of public and private organizations, arguing that it is not feasible to identify
distinctly private or public organizations.
While these foundational theorists suggest difficulty in drawing specific
distinctions between public and private organizations, more recent research has attempted
to clarify the differences. The focus on which type of organization can produce better
outcomes (and in what circumstances) grew in response to a major effort by governments
in the United States, Europe, and elsewhere to decentralize government authority and
privatize government services under the heading of New Public Management (Hood,
1991). This emphasis on determining whether public or private organizations produce
better results is a key field of research in public administration, however, it starts with an
assumption that there are clear, consistent, and easily identifiable criteria to distinguish
the two types of organizations.
What makes an organization public or private? A significant vein of the
literature has focused on the criteria that differentiate organizations in the two sectors.
Doing so can be difficult and researchers have used different approaches, including
placing organizations on a public-private continuum rather than drawing strict
distinctions (Bozeman, 1987; Moulton, 2009; Rainey, Backoff, and Levine, 1976; Rainey
& Bozeman, 2000). Criteria can include: involvement (or not) in the economic market for
revenues and resources; the extent of formal legal constraints placed upon the
organization; the intensity of outside political influence on the organization; the extent of
the organizations ability to coerce behavior; the scope of impact of policy decisions by
the organization; the extent of public scrutiny and public expectations; the type of goods


37
produced (public, quasi-public, or private); the criteria for evaluation; the extent of
fragmented authority; the emphasis placed on performance; and the incentives offered to
workers (Rainey, Backoff, & Levine, 1976). A more succinct set of criteria focuses on
who owns the organization, how it is funded, and what type of social control is exerted
upon it (Bozeman, 1987; Hvidman & Andersen, 2013; Perry & Rainey, 1988; Walmsley
& Zald, 1973). However, even these relatively straightforward criteria can lead to
classifying the same organization differently (Meier, OToole, & Hicklin, 2009). To
further simplify the concept, some researchers have used the relatively straightforward
criterion of regulatory authority and ownership to determine whether an organization is
public or private (Bozeman & Bretschneider, 1994; Feeney & Welch, 2012; Gibson,
2011; Monks, 2000).
Thus, throughout public management literature, there are a variety of operational
definitions for determining what makes an organization public or private. The field has
not reached consensus on the best way to approach the topic. However, research focused
on certain sub-fields, such as postsecondary education, has identified usable distinctions
between public and private organizations.
Public and private in higher education. Determining what exactly a public or
private college or university would be somewhat difficult based on the numerous
criteria delineated above. Based on the definition of whether or not an organization
pursues public purposes, colleges and universities could all be considered public
(Bozeman, 2013; Meier & OToole, 2011; Meier, OToole, and Hicklin, 2009). Using
main sources of revenue could also be problematic. Although nominally public colleges
and universities all receive some government funding, some research suggests that those


38
that have high tuitions may behave more like private colleges and universities,
particularly when they operate in a decentralized state governance regime (Knott and
Payne, 2004). Further, as direct governmental funding of publicly-owned colleges and
universities has declined, research suggests they have become more like privately-owned
schools (Duderstadt and Womack, 2003). Additionally, given that many privately-owned
colleges and universities derive significant revenue from tuition paid by federal grants
and loans, as well as receive significant public revenue in the form of federal research
grants, they certainly could be considered public entities under some of the criteria above
(Meier, OToole, & Hicklin, 2009).
Within higher education policy and research circles, however, colleges and
universities are regularly classified as public or private based on political control and
funding essentially the concept of ownership used elsewhere in public management and
higher education literature and research (see for example College Board, 2013; Feeney &
Welch, 2012; Meier, OToole, and Hicklin, 2009; Monks, 2000; National Center for
Education Statistics, 2013; Tierney, 1980). Schools that are overseen by public agencies
or entities and funded directly by state governments are classified as public, while those
that are not are classified as private. Within the label of private, there are for-profit and
non-profit schools, which are categorized based on their legal status (Cellini &
Chaudhary, 2014; Deming, Goldin, & Katz, 2011). These classifications are incorporated
in the Carnegie classifications, which are used throughout the higher education policy
n
and research communities. While these definitions ignore some of the nuance identified
elsewhere in studies on publicness, they do provide fairly straightforward categorization
that is amenable to empirical research. Further, adhering to consistent definitions across 7
7 For additional information on the Carnegie Classifications, please see http://carnegieclassifications.iu.edu/


39
research efforts is crucial for developing meaningful and comparable results (Meier,
OToole, &Hicklin, 2009).
Publicness as an independent variable. Research looking at the outcomes of
public, private non-profit, and private for-profit schools must address issues of selection
bias. Studies show that the student populations vary significantly by sector. Students at
private for-profit schools are more likely to be racial/ethnic minorities, tend to have
lower-incomes upon entering, and are more likely to be adults than students attending
schools from one of the other sectors (Cellini & Chaudhary, 2014; Deming, Goldin, &
Katz, 2011). Other research disagrees, showing that students at public colleges tend to
have lower incomes and are more likely to be adults and part-time students than those at
private schools; the contradiction may be due to the fact that this research does not
capture recent growth in the for-profit sector (Scott, Bailey, & Kienzl, 2006).
Research using this operational definition of publicness includes studies that
examine a variety of dependent variables, including student demand for higher education,
output by faculty, and overall economic returns to students (Feeney & Welch, 2012;
Monks, 2000; Tierney, 1980). There are also studies on graduation rates that show a
range of findings, including that public schools outperform private non-profit ones, that
there is no significant difference, and that private non-profit schools outperform public
ones (Gibson, 2011; Meier, OToole, & Hicklin, 2009; Scott, Bailey, and Kienzl, 2006).
None of this research examines privately-owned for-profit schools. Although federally-
reported graduation rates for private schools are, on average, higher, the strategy for
controlling for selection bias impacts findings (Scott, Bailey, and Kienzl, 2006). One
important variable seems to be the amount of funding available per student. When it is


40
included as a control variable, public schools outperform private ones (Scott, Bailey, and
Kienzl, 2006). Student characteristics and controls for ability bias also impact findings
(Gibson, 2011; Meier, OToole, & Hicklin, 2009; Scott Bailey, and Kienzl, 2006). Most
of this research, due to its reliance on federal data collected on first-time, full-time
students, explicitly excludes near completers.
Other research on the connection between student outcomes and the type of
college or university attended has focused in recent years on the differences between
outcomes for graduates of private for-profit schools compared to graduates from private
non-profit and public schools. Examining outcomes, graduates from private for-profit
schools have higher loan default rates, lower earnings, and higher unemployment rates
compared to graduates from other schools (Cellini & Chaudhary, 2014; Deming, Goldin,
& Katz, 2011). Other research has shown students completing two-year degrees at for-
profit schools have slightly larger economic returns than those from non-profit schools,
but this may be due to selection bias because many students who attend public and non-
profit two-year schools subsequently pursue four-year degrees which may negatively bias
their immediate earnings after graduation (Lang & Weinstein, 2013).
Additional research combines measures of school quality with sector type.
Research examining outcomes based on selectivity of the college or university and its
public or private non-profit status finds large economic returns for graduating from a
highly selective private non-profit school compared to a low-rated public school (Brewer,
Eide, & Ehrenberg, 1999). Comparing public and private non-profit schools of the same
category shows no statistically significant earnings differential (Zhang, 2005). However
research using more effective methodologies to control for selection bias, such as


41
matching similar students who attended different types of colleges and universities, as
well as different quality levels, finds little difference in wages earned by graduates (Dale
and Krueger, 2002). Other research shows slight earnings gains for white students who
finish degrees at private non-profit colleges and universities compared to public schools,
and no statistically significant earnings differential for non-white students (Monks, 2000).
In research examining gender differences, Joy (2003) finds that women who attended
private non-profit schools that granted doctorates (a rough measure of quality) earned
more than those who attended other private non-profits and those that attended public
colleges of either type (Joy, 2003). For males, there was no statistically significant
difference (Joy, 2003).
While these results show disagreements within the field of higher education
research, they do not provide much guidance for the study at hand because they generally
focus on traditional students and exclude near completers. However, this research is
informative because it shows that accounting for the characteristics of the students at
schools in different sectors is crucial for understanding any differences in outcomes due
to public, non-profit, or for-profit control.
Thus, while there is consensus in the literature on the criteria used to differentiate
public, non-profit, and for-profit colleges and universities, there is not agreement about
earnings differentials by school management type. Following the overwhelming majority
of research into differences between outcomes at different types of colleges and
universities, I use control of the institution, as defined by the widely-accepted Carnegie
classifications. Although there is limited research that focuses on near completers who
return to finish a degree and how the choice of school may affect their outcomes, research


42
does generally show that there may be differential outcomes due both to student and
school characteristics.
Literature Review: Conclusion
This review of literature spans many different topical areas. While there are few
directly relevant studies focusing on near completers, research from labor economics and
public management has identified numerous critical issues for conducting a study of the
economic return for near completers who finish degrees. Given the many different topics
covered in the literature review, and the lack of a deep body of research on near
completers, it is helpful to highlight the important conclusions. Methodologically, the
immense body of literature on the returns to education is nearly unanimous on the
importance of accounting for potential selection bias. Bias due to ability as well income
before and during enrollment could affect results if not properly controlled. Further,
although studies may reach opposing conclusions on the benefits to certain sub-groups,
there is ample evidence that returns to education are not uniformly distributed by
racial/ethnic groups and gender. Additionally, research on outcomes when services are
provided by public, non-profit, and for-profit sectors are not conclusive, especially in the
education realm. Given the variations evident when looking at some outcome measures,
further study is warranted with regards to near completers. The research cited above
informs the hypotheses I generate in the subsequent section, as well as the
methodological approaches selected in Chapter 4.


43
Hypotheses
This dissertation will contribute to policy discussions, to the academic literature
on human capital and signaling theories, and to literature on the differences between
public and private organizations by evaluating the four hypotheses. These hypotheses are
collected in Table 4 and linked to the original research questions that guide this
dissertation
Hypothesis 1.1: Near completers who return to postsecondary education to finish
a baccalaureate degree will earn a positive individual wage premium compared to those
who do not return. The premium is sufficiently large to result in a positive economic
return. The literature cited above supports the conclusion that, on average, earning a
bachelors degree results in positive wage premiums. Those premiums tend to be
substantively large enough that most research does not bother accounting for direct and
indirect costs of attaining the degree. Given that near completers are much more likely to
be older than students who proceed directly through postsecondary education, they will
have less time in the labor market to recoup costs. Also, near completers who likely
earn higher wages than students proceeding directly from high school through college -
may face higher indirect costs due to foregone wages. Additionally, with data showing
that a substantial percentage of adults enroll in privately-owned for-profit colleges and
universities that cost more than public schools, it could be that, on average, returning near
completers are paying higher direct costs than other types of postsecondary students,
resulting in lower economic returns. Even with conservative estimates for direct costs, it
is likely that at least some subgroups earn a positive economic return for baccalaureate
degree completion.


44
Hypothesis 1.2: The benefits of degree completion by near completers do not
accrue equally to all population sub-groups. The literature cited above suggests that there
may be heterogeneous returns to education attainment based on subgroups disaggregated
by gender, racial/ethnic background, and socioeconomic status. It is likely that these
differences will also appear in an analysis of near completers.
Hypothesis 2: Individuals who finish a degree at a public college or university
will realize a higher economic return than those who finish a degree at a non-profit or
for-profit college or university. While the literature is mixed on the differential returns to
traditional students based on sector, data do show that non-profit and for-profit colleges
and universities cost more than in-state tuition at public schools. In order to have the
same overall return as those who attend lower-priced public colleges, near completers
who finish baccalaureate degrees at non-profit and for-profit schools likely must earn a
higher wage premium for degree completion to offset the higher direct costs. However,
given that there are not data on actual tuition prices paid by any individual, and that
access to various forms of financial aid can vary significantly, it is important to
understand that these approximated returns are based heavily on a series of assumptions
about direct costs. An additional factor that could challenge this hypothesis is that for-
profit schools tend to emphasize the flexibility of their programs for adults, which could
mean that graduates incur lower indirect costs in the form of foregone wages. Unless the
results show higher wage premiums (or lower indirect costs) for graduates of non-profit
and for-profit schools, it is likely that their overall return will be lower. Granted, such
conclusions would be somewhat tentative due to the underlying assumptions that sources


45
of financial aid do not vary by sector and actual tuition price paid by students is
comparable to the average prices cited below.
Hypothesis 3: Factors that are important in predicting educational successes for
traditional students will also be important in predicting whether near completers finish
degrees. Research on educational attainment suggests a number of factors that impact the
level of education an individual completes. These include socio-economic status,
racial/ethnic background, family characteristics, as well as unobservable characteristics
such as motivation and ability. It is likely that many of the factors that impact traditional
students educational attainment will also apply to near completers.
Table 4: Research questions and hypotheses
Research Question Hypotheses
RQ1: Do near completers who return to finish a degree earn a positive economic return compared to near completers who do not return? H1.1: Near completers who return to postsecondary education to finish a baccalaureate degree will earn a wage premium compared to those who do not return. The premium is sufficiently large to result in a positive economic return.
H1.2: The benefits of degree completion by near completers do not accrue equally to all population sub-groups.
RQ2: Does the type of management at the college or university from which a near completer graduates affect his/her economic return? H2: Individuals who finish a degree at a public college or university will realize a higher economic return than those who finish a degree at a non-profit or for-profit college or university.
RQ3: How do the characteristics of individual near completers affect the likelihood that they will return to a college or university and complete a degree? H3: Factors that are important in predicting educational successes for traditional students will also be important in predicting whether near completers finish degrees.


46
Contributions to the Field
Drawing on the limitations identified in that literature review regarding research
specific to near completers, this section identifies ways in which evaluating these
hypotheses will contribute to the field both empirically and theoretically.
Theoretical development: Human capital and signaling theories. Although
there does not appear to be any extant rigorous research specifically focusing on the
research questions posed above, one could argue that human capital and signaling
theories suggest there will be a positive individual economic return for near completers
who finish degrees. However, several characteristics of near completers make it difficult
to reach that conclusion based on these theoretical frameworks. First, near completers are
older than traditional students that formed the basis of the original theories and thus have
a shorter window in which to recoup costs of education. Second, when these potential
students return and finish a degree, they are receiving a relatively limited amount of
additional education, depending on how close to earning a degree they are. Empirical
tests of the benefits of shorter term training programs for adults, though not completely
analogous to completing a college degree, have shown mixed results. Third, much
research in these areas has focused on fairly idealized cases that may approximate reality
for traditional-aged students, but the underlying assumptions on direct and indirect costs
may have more significant impacts on older students.
Research on these questions could help to address whether changes in an
individuals human capital stock later in life have a similar effect as changes early in life.
If one considers an individuals human capital stock to include education, experience, and
training, then it could be that the additional education later in life makes a smaller


47
contribution to the overall stock in relative terms, resulting in a smaller impact of
increases in education levels in the adult years. With near completers already possessing
significant prior college credit, it could be that the additional postsecondary schooling
does not offer significant economic benefits. Alternatively, completing a postsecondary
degree could be an important credential with tangible workplace benefits.
Signaling theory and the sheepskin effect, broadly speaking, would suggest that
near completers who finish a degree should receive an economic benefit. However, it
could be that for near completers who have been in the workforce for many years, work
experience supersedes the signal sent by their new diploma. In essence, their employers
could already have a fairly complete view of the difficult-to-observe characteristics that a
diploma represents as experience sends a stronger signal than a new degree.
Alternatively, it could be that earning a college diploma sends a new signal to employers
and opens up a variety of new, more lucrative opportunities. Related to Arrows (1973)
work, it could be that an individual miscalculated the amount of education he or she
needed to adequately match his or her characteristics, leading to underemployment.
Similarly, if an individuals innate characteristics, such as perseverance or work ethic
have changed over time perhaps through general maturity or due to life experience it
could be that the new credential better captures these new factors and sends a more
accurate signal to employers. This research effort will extends these theories to address
near completers by answering these different questions.
Outcomes of public and private organizations. This research will also make a
modest contribution to the stream of literature focused on whether management type
impacts outcomes. The straightforward definition of public and private in research on


48
higher education will limit that contribution to the subset of studies that use ownership
and/or political control as the operational definition. Within that subset, there is limited
research that examines whether management type may affect outcomes differently for
subpopulations such as adults. Within research on higher education, there are no apparent
studies that examine outcomes for near completers by the type of management.
Factors that affect the pursuit of additional education. Similarly, there is no
apparent research about the individual characteristics that may be associated with the
decision by near completers to return to postsecondary education to complete a degree.
There has been some market research testing which messages resonate with near
completers, but no academic research has examined the characteristics of those who
return to finish degrees compared to those who do not. This examination could help
better target policies and programs while also filling in gaps in research about decisions
of non-traditional students to pursue additional education.
Empirical contributions. The major empirical contribution that this research
makes to the field is relatively straightforward. There is little rigorous research specific to
whether near completers receive an individual economic benefit from returning to finish a
baccalaureate degree beyond program evaluations that offer little comparison and few
control variables. Policymakers (and potentially returning students) currently rely on
simple and straightforward bivariate comparisons of earnings by bachelors degree
holders and those with some college, but no degree to make the case for broad programs
aimed at these potential students.
Beyond this central contribution, there are numerous other empirical questions
that will be answered in the course of addressing the central question. As a beginning


49
point, there is little basic research on the characteristics of near completers. The some
college, no degree census category captures all those who have completed any college
credits, and data based on this category underpins most of the policy and programmatic
conclusions about near completers. Policy arguments to start degree completion programs
serving near completers have focused on the fact that 22 percent of the adult population
falls into this category (U.S. Census Bureau, 2011). However, state programs are focused
on a subset of this population those close to degrees and little is known about this
group. One side benefit of this research is estimates of the demographic makeup of near
completers.
Additionally, little is known about the demographics and personal characteristics
of this group. Although it would be relatively straightforward to conduct demographic
research on the broader category from the census, it is not certain that this would
accurately reflect the makeup of near completers. Given what is known about college
completion rates by race and gender, it could be that this racial/ethnic background and
gender are not evenly distributed in this population. Such information would be useful to
practitioners seeking to reach and reengage this group, to policymakers seeking to
address racial/ethnic differences in degree attainment, and to researchers examining
differential returns to education. Additionally, it could be that the credit distribution is
correlated with age. If near completers turn out to be much older than traditional students,
or those who finish degrees do so after many years away from postsecondary education,
it could complicate policy arguments that serving these individuals will address future
workforce needs because these individuals may not work for as many additional years.


50
Costs for near completers. A general critique of literature on the benefits of
education is that much of it focuses only on the wage premiums associated with greater
attainment while ignoring costs. This is partly due to the assumptions ingrained in
Mincers foundational work on estimating returns to education. Further, even when
accounting for more realistic conditions, such as costs of tuition, most estimates of the
wage premium for college degrees would suggest a positive return on investment even
with extremely conservative estimates of future earnings and overestimates of costs
(Hout, 2012). Estimates of the economic return on college degrees almost uniformly
show that the benefits substantially outweigh the costs, with some notable differences by
racial/ethnic subgroup or gender (Day & Newburger, 2002; Hout, 2012; Julian &
Kominski 2011; Kane & Rouse, 1995). Research has not specifically focused on the cost
and potential returns for near completers, who have a shorter window in which to recoup
education costs. Compounding this, if near completers have to forego earnings to pursue
additional education, their indirect costs may be much higher than has been estimated for
traditional students because they likely have much higher salaries than traditional-aged
students, although this would be offset if their subsequent wages are also
commensurately large.
Estimating foregone earnings for near completers is an important addition to the
field. As noted earlier, some research on adults who earn degrees and certificates finds
lower pre-treatment earnings, but little research focuses on whether those who do pursue
additional education and training sacrifice wages to do so (Ashenfelter, 1978; Blanden, et
al., 2012; Jepsen & Montgomery, 2012; Marcus, 1986; Light, 1996). Conclusions based
on this evidence differ, with some arguing that lower incomes for adults who enter


51
training and education programs means that opportunity costs and foregone earnings are
barriers to adults pursuing further education, and others arguing that those with lower
incomes may have greater economic needs and thus more reasons to enter school or a
training program (Blanden, et al., 2012; Jepsen & Montgomery, 2012; Marcus, 1986).
Further research will help identify how earnings change prior to and during additional
education. One might also expect that, especially with the flexible adult-focused
programs being offered by institutions of higher education, few adults who are already
employed and have the normal financial obligations of life would be likely to quit their
jobs and embark on a traditional education program full-time with classes during the day,
so their economic indirect costs may be less than those assumed by the standard model
proposed by Becker. Alternatively, it could be that the growth in on-line learning and
other more flexible options is too recent to be captured by the dataset used here or that
any commitment to complete a degree requires the student to forego some work
opportunities.8
Measuring each individuals actual direct costs for college attendance is not
feasible even with information about the school that he or she attended, due to financial
aid packages and tuition discounting that affect the actual price paid. The analyses used
in this research do not account for grants, scholarships, and tuition discounts, mainly
8 A true accounting of indirect costs includes not just foregone wages, but the cost of other non-economic
losses that an individual might face in returning to complete a degree. These include a loss of free time,
spending less time with family, and the psychic costs associated with pursuing additional education. This
study focuses exclusively on economic indirect costs. For additional discussion of non-economic indirect
costs see Heckman, Lochner, and Todd (2006).


52
because such data are not available, but given an appropriate data source this would be a
valuable area for additional research.9
Contributions to the field: Conclusion This research will make a significant
contribution to human capital and signaling theories, while also adding to literature on the
differences between public and private organizations. Additionally, this research will
contribute to ongoing policy discussions about future workforce needs and degree
completion programs. Although the policy community already explicitly assumes certain
results that this effort purports to provide, as I have shown through this literature review,
many of the assumed results are not yet supported by solid empirical results. This
research will provide important theoretical development and empirical evidence to either
support or alter existing national, state, and local policies, while also providing some
clarity about differences in performance of colleges and universities by management
type.
9 Tuition discounting the phenomenon in which schools do not charge full price likely has a significant
effect on estimated costs. Abel and Dietz (2014) provide additional discussion in the context of estimating
returns to education.


53
CHAPTER III
DATA AND MEASUREMENTS
This section describes the data source employed to test hypotheses proposed
above. I present a wide range of descriptive data and illustrative figures that inform the
chosen methodology, choice of variables, and approaches for producing estimates to test
my hypotheses. Most of the descriptive data tables employ survey weights to present a
representative picture of near completers.
Data Source
To test the hypotheses posed above, I use the National Longitudinal Survey of
Youth 1979 (NLSY). This longitudinal survey of 12,686 individuals has been conducted
annually from 1979 to 1994 and biannually from 1994-2012. All survey respondents
were between ages 14 and 22 when the survey commenced in 1979. The survey is
designed to be nationally representative, but is composed of three sub-samples: the first is
6,111 nationally representative individuals; the second subsample is an oversample of
5,295 black, Hispanic, and low-income non-black, non-Hispanic individuals; the final
subsample is a group of 1,280 individuals enlisted in the military at the time of the initial
survey (Bureau of Labor Statistics, n.d.).
Sample description. Within the overall NLSY sample, I use 1059 individuals
who attained the status of near completer at some point during the survey. I
operationalize this term as an individual who finished at least half of a baccalaureate
degree without completing it, but then has at least one survey round in which he or she


54
was not enrolled in college.10 As this is the central definition for this study, additional
justification and analysis of this operationalization is warranted.
Due to the limited extant research on this topic, there is not a widely accepted
operational definition of near completer to employ. Shapiro, et al. (2014), use two
years of enrollment as marking significant progress toward a degree in delineating a
group they call potential completers, but this is more of a general report examining the
number of students meeting these criteria. Of the other studies purporting to examine
returning students, there are few usable operational definitions. Hoyt and Allred (2008)
evaluate a program that targeted students who had completed 30 credits (approximately
one full-time year of enrollment) who had not been enrolled in two years, while other
research is not specific about the definition.
The varied programs targeting these students that are currently in operation
employ a range of definitions as well. A list of definitions from such programs is
included in Table 5, though it is far from comprehensive. Although there is not a
consensus, one key part of the definitions used is non-enrollment. There is much
variation in the cut-off for the number of credits that marks substantial progress toward a
degree. The definitions vary from any credits at all to 75 percent of the way toward a
degree. Using 50 percent of the credits necessary for a degree represents middle ground
among these varied definitions.
10 This measure is based on years of postsecondary completed, which is derived from the highest grade
completed variable. For postsecondary education, each year represents the traditional years of college.
An individual who attended college for two years part-time but only earned credits necessary to finish one
quarter of a degree would have a high grade completed of 13 even though he or she completed two calendar
years in postsecondary education.


55
Table 5: College and state definitions for degree completion programs
State Definition Source
Arkansas 75 percent of credits necessary for a degree, at least 22 years old, out of college at least two years. Lane, Michelau, & Palmer (2012)
Brigham Young University At least 30 credits (25 percent of those necessary for a degree), not enrolled for two years. Hoyt & Allred (2008)
Colorado 75 percent of credits necessary for a degree, at least 25 years old, no longer enrolled. Lane, Michelau, & Palmer (2012)
Connecticut At least 12 credits completed. At least 18 months of non- enrollment. Connecticut House Bill 5050 (2014)
Kentucky 2/3 of the credits necessary for a degree, not enrolled. Kentucky Council on Postsecondary Education (2008)
Maryland At least 90 credits competed, not enrolled. Maryland Senate Bill 0740 (2013)
Minnesota Some college credits (not specified) but no degree, not enrolled. Minnesota State Colleges and Universities (2015)
Nevada No credit limit, has not been enrolled in previous year. Lane, Michelau, & Palmer (2012)
New Jersey Credit definition left up to schools, not enrolled Lane, Michelau, & Palmer (2012)
North Dakota Completed 70 percent of credits, not enrolled Lane, Michelau, & Palmer (2012)
Oklahoma 72 credits, at least 21 years old, not enrolled, completed general education requirements. Oklahoma State Regents for Higher Education (n.d.)
South Dakota 90 or more credits, not enrolled Lane, Michelau, & Palmer (2012)


56
Other operational definitions are plausible, as well. I consider the following alternatives.
1. Individuals who complete at least 50 percent of a baccalaureate degree and then
make no further progress on their degree for three years. The goal of the second
part of the definition is to ensure that the individual is no longer enrolled in
postsecondary education. While this slow progress could indicate that, it could
also be that some of these individuals are progressing slowly while enrolled,
likely due to being part-time students. Given that individuals report their
education status during each round, for the years in which the survey takes place
every two years, it is not possible to determine a three year gap with precision.
Thus, the operational definition is that an individual remains at the same level of
education for three consecutive survey rounds for 1979-1994 and two consecutive
survey rounds from 1994-2012. The latter years effectively become a pause in
progression of at least four years.
2. Individuals who complete at least 50 percent of a baccalaureate degree and have a
spell of non-enrollment that lasts at least two years. This definition accomplishes
the main goal of ensuring that an individual is no longer enrolled, but similar to
previous definition, it is not clear how this definition improves on the preferred
one. The only difference is in the length of time away from earning a degree.
Considering only the years 1994-2012, this results in an identical population as
the main definition due to the biennial nature of the survey during these later
years. For the years 1979-1994, it eliminates from consideration those individuals
who were not enrolled for only one survey round.


57
3. Individuals who complete at least 75 percent of a baccalaureate degree and then
report a spell of non-enrollment.
From a policy perspective, if the goal is to identify those individuals who have completed
significant credits and can complete degrees relatively quickly, selecting a lower cut-off
does not seem appropriate as those who are less than 50 percent of the way towards a
degree would still be technically in lower division classes. The descriptive data presented
below, as well as the results and discussion that follow rely on the operational definition
chosen above. I include sensitivity tests comparing results using the potential alternative
definitions and a further discussion of the implications for this operational decision in the
appendix.
Descriptive data of the background characteristics of these individuals who
attained near completer status, as defined above, are presented in Table 6. The data show
that the average age was just under 18 when individuals in the sample were first
interviewed. Parents of the sample respondents completed between 12 and 13 years of
formal schooling on average, and 12 percent of the sample fell below poverty line in
1979. The survey also collected data on respondents scores on the Armed Services
Vocational Aptitude Battery (ASVAB), which is a standardized test for qualification for
the armed forces frequently used in literature on education and earnings as a control for
ability (Black & Smith, 2004; Monks, 2000; Kane & Rouse, 1995; Taniguichi &
Kaufman, 2005). Survey respondents took the ASVAB at the outset of the survey and
their ages ranged from 14-22 when taking the exam, which would bias the scores
upwards for older individuals and downwards for younger individuals. I adjust these
scores rescaling them based on the individuals age when taking the exam. This


58
Table 6: Descriptive statistics for those who attained near completer status
Variable Mean Linearized Std. Err. Min Max
1979 Age 17.72 0.10 14 22
Magazines in the home (1979) 0.76 0.02 0 1
Newspapers in the home (1979) 0.88 0.01 0 1
Possessed library card (1979) 0.83 0.01 0 1
Mothers high grade 12.45 0.10 0 20
Fathers high grade 12.84 0.15 0 20
Family below poverty line (1979) 0.12 0.01 0 1
ASVAB Score (1979) 68.08 0.82 3 99
Male 0.48 0.02 0 1
Race/Ethnicity
Hispanic 0.06 0.01 0 1
African American 0.16 0.01 0 1
Non-African American/Hispanic 0.77 0.01 0 1
N 1059
Note: Data are weighted using sampling weights provided by NLSY, with errors
adjusted to account for stratified cluster sampling strategy.
procedure follows Monks (2000) and Taniguichi and Kaufman (2005). A small portion of
the sample (<4%) did not report ASVAB scores. I follow Light and Strayer (2004) and
assign these individuals the mean score.11 The weighted average for those individuals in
the subsample is just above the 65th percentile. Just under half of the subsample is male,
and the race/ethnicity backgrounds of near completers is 77 percent non-African
American/Hispanic, 16 percent African American, and 7 percent Hispanic.
Survey attrition and changes. Although the survey is widely used and accepted
in social science research, there have been a number of changes in the NLSY79 over time
that could potentially affect findings and must be acknowledged. First, as in all
longitudinal surveys, non-random attrition is a concern. Between 1979 and 1994, the
11 Dropping those individuals with missing ASVAB scores from the analyses presented later in this paper
does not substantively affect results when compared to including them in the analyses that rely on ASVAB
scores.


59
overall retention rate was 89 percent and 75 percent of the individuals from the main
sample who started the survey in 1979 responded in 2010 (Bureau of Labor Statistics,
n.d.). The overall attrition rate is relatively small compared to other longitudinal surveys
over such a time period (Hernandez, 1999). More recent analysis shows that the NLSY
reinterview rate averages about 96 percent, which compares favorably to other large
longitudinal surveys (Schoeni, Stafford, McGonagle, & Andreski, 2013). There is limited
evidence of non-random attrition, as shown in Table 7, which shows unweighted
descriptive data for the sample used in 1979 compared to the portion of that sample that
was also interviewed in 2012.
The most striking difference between the 1979 and 2012 samples is the decrease
in the proportion of the sample that was in poverty in 1979. While a portion of this may
be due to standard attrition, it is likely that the bulk of this change comes from the
decision by survey administrators to drop part of the low-income oversample in 1990
(Bureau of Labor Statistics, n.d.). Researchers have examined attrition in the NLSY and
conclude that the relatively small differences are not likely to bias estimates (Sen, 2006).
Further, NLSY79 is widely used for research in a variety of subject areas and findings
based on the data are generally accepted in academic literature (MaCurdy, Mroz, & Gritz,
1998).
Still, it is possible to adjust the data using survey weights provided by NLSY to
correct for attrition in addition to correcting for oversampling. Repeating the descriptive
data with appropriate survey weights applied to the subsamples of interest minimizes the
differences due to attrition, as can be seen in Table 8. Adjusted Wald tests show that there
are no significant differences in the means for any of the variables is this table.


60
Table 7: Changes in descriptive data (unweighted) 1979 to 2012
Variable Mean 1979 SD 2012 Mean SD
1979 Age* 17.91 2.35 17.63 2.24
Magazines in the home (1979) 0.68 0.47 0.66 0.48
Newspapers in the home (1979) 0.82 0.39 0.79 0.41
Possessed library card (1979) 0.79 0.41 0.78 0.41
Mothers high grade 11.82 3.16 11.73 3.16
Fathers high grade 12.05 3.95 11.86 3.96
Family below poverty line (1979)* 0.21 0.41 0.20 0.40
ASVAB Score (1979) 61.86 24.79 59.33 25.66
Male 0.46 0.50 0.44 0.50
Race/Ethnicity
Hispanic 0.16 0.37 0.18 0.38
African American 0.29 0.46 0.36 0.48
Non-African American/Hispanic 0.54 0.50 0.46 0.50
N 1059 648
Denotes a statistically significant difference in means at p<05.
Table 8: Weighted descriptive data, 1979 and 2012
Variable 1979 .. Linearized Mean Std. Err. 2012 .. Linearized Mean Std. Err.
1979 Age 17.72 0.10 17.85 0.11
Magazines in the home (1979) 0.76 0.02 0.75 0.02
Newspapers in the home (1979) 0.88 0.01 0.87 0.01
Possessed library card (1979) 0.83 0.01 0.82 0.02
Mothers high grade 12.45 0.10 12.35 0.13
Fathers high grade 12.84 0.15 12.71 0.18
Family below poverty line (1979) 0.12 0.01 0.13 0.02
ASVAB Score (1979) 68.08 0.82 67.32 1.02
Male 0.48 0.02 0.49 0.02
Race/Ethnicity
Hispanic 0.06 0.01 0.07 0.01
African American 0.16 0.01 0.17 0.01
Non-African American/Hispanic 0.77 0.01 0.77 0.01
N 1059 648


61
Educational attainment variables. An individuals educational attainment is the
key variable in this analysis. As noted above, near completer is operationalized as an
individual who completed more than half of a baccalaureate degree without completion
and had a period of non-enrollment. This category excludes those who earned associates
degrees, or attained near completer status then subsequently earned an associates degree.
Although associates degrees (and postsecondary certificates) are important for
understanding the broader credential completion environment, they are beyond the scope
of this study. Further research into these types of credentials, and their impact on
individual wages and social outcomes such as filling workforce demand is certainly
warranted. This research is intended to focus specifically on outcomes for near
completers who finish baccalaureate degrees, which could be confounded by including
those who earn both bachelors and associates degrees.
Within the subset of individuals who attain the status of near completer at some
time during the survey, respondents are divided into four categories in any year after they
first attained near completer status: 1) near completers, operationalized as described
above but who do not fall into one of the following three categories; 2) near completers,
pre-enrollment, which includes those individuals who are near completers and who
report enrolling in postsecondary education in the subsequent interview period; 3) near
completers, enrolled, which includes those individuals who are near completers who
report being enrolled in college; and 4) near completer, finished degrees, which
includes those individuals who were near completers at one point but returned and
completed a baccalaureate degree.


62
The second category is included to control for the Ashenfelter Dip where
income declines just prior to an individual deciding to pursue additional education or
training. The third category has two purposes. First, including this variable controls for a
temporary drop in income caused by foregone wages that could bias estimates of the
treatment effect in the same way as the Ashenfelter Dip. Second, estimating this
coefficient will provide important information about indirect costs for calculating the cost
of attending college for near completers.
Collapsing those four categories to just two for the purposes of providing
descriptive data shows how the population of near completers and those with degrees
changes over time. The changes over time in the number of near completers and those
near completers who finished degrees are displayed in Figure 1. The number of near
completers grows rapidly through 1987, at which point the growth starts to level off. The
number of near completers who finish degrees continues to grow slowly but steadily
throughout the rest of the survey. This initial analysis ignores attrition, and once an
individual attains near completer status, he or she is counted in the categories above in
perpetuity, regardless of whether he or she responded to the survey in a given year.
Accounting for attrition shows slight differences, as can be seen in Figure 2.


63
Figure 1: Near completers and near completers finishing degrees over time
Figure 2: Near completers and near completers finishing degrees over time,
accounting for attrition
This second figure shows a substantial decline in the near completer population
between 1989 and 1990, likely reflecting the changes to the sample and the decision by
administrators to discontinue interviews with certain subsamples. Following that large
drop, the number of near completers declines slightly through the rest of the survey. The
number of near completers who finished degrees grows somewhat quickly through 1990
(though less rapidly than the population of near completers), before its growth rate
decreases slightly, growing slowly through the rest of the survey. Some of this growth


64
likely contributes to the declines in the population of near completers who did not finish
degrees.
Which near completers finish degrees? One key question for this research effort
is about the individual characteristics that may predict whether a near completer returns
to finish a degree or not. Comparing those who attained the status of near completer but
never finished a degree with those near completers who at some point finished a
baccalaureate degree shows notable statistically significant differences, as can be seen in
Table 9. Several characteristics that are associated with larger educational attainment
Table 9: Demographic data by educational attainment
Variable Near Completers, Never Finished .. Linearized Mean Std. Err Near Completers, Finished Degrees .. Linearized Mean Std. Err
1979 Age 17.75 0.12 17.67 0.17
Magazines in the home (1979)*** 0.72 0.02 0.82 0.02
Newspapers in the home (1979)** 0.86 0.01 0.91 0.02
Possessed library card (1979) *** 0.79 0.02 0.89 0.02
Mothers high grade*** 12.06 0.12 13.14 0.18
Fathers high grade*** 12.25 0.17 13.82 0.25
Family below poverty line (1979)*** 0.14 0.01 0.08 0.02
ASVAB Score (1979)*** 63.72 1.02 75.71 1.26
Male 0.48 0.02 0.48 0.03
Race/Ethnicity
Hispanic 0.08 0.01 0.03 0.01
African American 0.19 0.01 0.12 0.01
Non-African American/Hispanic 0.73 0.02 0.85 0.01
N 740 319
Denotes a statistically significant difference in means at p<10
** Denotes a statistically significant difference in means at p<05
Denotes a statistically significant difference in means at p<01
Denotes a statistically significant difference in distribution at p<01 using a corrected
Pearsons chi square test.
Data are weighted using survey weights provided by NLSY.


65
elsewhere in the literature show significant and substantively large differences, including
parental education, receipt of magazines and newspapers, possession of a library card,
and ASVAB scores,which all tend to be higher among those who finish degrees. Initial
poverty levels tend to be lower among degree finishers. The racial and ethnic distribution
of the two groups also appears to be different, with a higher percentage of individuals
who are neither Hispanic nor African-American among degree finishers. The difference
in the distributions is statistically significant.
This initial analysis is suggestive of some important differences between those
who finish degrees and those who do not, but estimating the impact of different variables
on whether an individual completes a degree will require an event history analysis. This
methodology will be described in detail below, but essentially this approach can examine
the event of interest (returning to complete a baccalaureate degree, in this case) over time
and estimate the impact of different time varying and time invariant characteristics on the
likelihood that an individual will earn a degree. This methodology also lends itself to
some useful descriptive statistics and graphics presented below.
The initial descriptive analysis, shown in Figure 3, is a Kaplan-Meier survival
estimate for near completers. Owing to this methodologys roots in biostatistics,
survival here means that the event of interest has not occurred and the individual
remains a near completer rather than a degree holder. The X-axis shows the length of
time in years and the Y-axis estimates the proportion of the population that remains near
completers rather than becoming near completers who graduated. Because survival is 12
12 One of the classic uses of event history analysis is the examination of the effect of medical treatments on
a patients subsequent survival.


66
equated with not graduating, the farther the curve drops below one, the more near
completers that have graduated.
This figure shows a curve that flattens out over time, suggesting that most near
completers who finish degrees do so within the first 10 years of attaining that status.
Additionally, the curve is even steeper in the first four years, suggesting that the near
completers are less likely to finish degrees as time passes. Part of this is likely due to how
college credits for reenrolling students are treated as former students try to reenroll.
Those who transfer to new schools tend to lose credits as their new school may not accept
all of the previously earned credits, or may accept them for elective credit only, forcing
the transfer student to retake classes that count towards a major, contributing to lower
graduation rates (Government Accountability Office, 2005). Additionally, some schools
may have policies affecting whether credits are accepted depending on how old they are.
Although schools may agree to accept all credits, regardless of age, it is often up to
individual departments to determine whether older credits are accepted for major credit,
or electives, again forcing returning students to retake classes they have already


67
completed. There is little research or policy analysis of this treatment of credits, but it is
an important consideration for returning students who may be set back several semesters
by enrollment policies at the schools they attend.
This Kaplan-Meier analysis can also be disaggregated by key variables. Figure 4
shows the same survival curve disaggregated by gender. The figure shows slight
differences in the survival curves for men and women, with men slightly more likely to
earn degrees in the first 15 years of being a near completer, while women overtake men
later in the study period. However, the differences between the two estimates are not
statistically significant using a log-rank test.
Kaplan-Meier survival estimates
Figure 4: Survival estimates for near completers by gender
Figure 5 shows a similar analysis for different racial/ethnic groups. These
estimates clearly show different survival rates for near completers of different
racial/ethnic backgrounds, with Hispanic near completers appearing less likely to finish


68
-1
Kaplan-Meier survival estimates
0 5 10 15 20 25 30 35
analysis time
Hispanic ----- African-American
........ Other ethnicity
Figure 5: Near completer survival rates by racial/ethnic group
degrees than African-Americans and individuals of other racial/ethnic backgrounds. The
difference between the latter two groups is more complicated with African-Americans
falling behind those from other racial/ethnic backgrounds initially, but then closing the
gap over time, which suggests there may be differences about when individuals from
different racial/ethnic backgrounds return to finish degrees. A log-rank test shows that the
differences between these survival curves are statistically significant. These results are
suggestive that there may be differences in the likelihoods for near completers to finish
degrees by racial/ethnic background.
I also examine survival rates by a familys poverty status in 1979. The Kaplan-
Meier estimates are presented below in Figure 6. This shows that those near completers
whose families were in poverty in 1979 are less likely to finish degrees over the course of
the survey. The differences between the two curves are statistically significant at p< 01.


69
Kaplan-Meier survival estimates
Figure 6: Near completer survival rates by familial poverty status
Additionally, across all of these Kaplan-Meier estimates, it is clear that the rate at which
near completers finish degrees changes over the length of time since they have attained
that status.
Income and degree completion. While there are clearly changes within the
sample in the education status levels of individuals, one important question for this
analysis is whether degree completions lead to changes in income levels. Income is
clearly dependent on numerous factors beyond just educational attainment, but examining
average incomes of near completers compared to near completers who finished degrees
can still be instructive. The Table 10 examines income differences between the two
categories at three points in time.
These data show that those near completers who finished degrees have earned
higher wages with strong statistical significance for all three years considered.
Additionally, the gap appears to widen over time, with the premium being approximately
one third of non-finisher wages in 1992 rising to slightly more than have of non-finisher
wages by 2012. Given the large number of other variables that may also impact income


(and could be correlated with education status), a more detailed multivariate analysis is
required.
70
Table 10: Income by educational attainment 1992, 2002, & 2012
Year Near Completers, No Degree Mean Linearized ^ Income Std. Err. Near Completers, Finished Degrees Mean Linearized ^ Income Std. Err.
1992*** $36,160 1,600 414 $49,914 4,821 139
2002*** $51,008 2,888 381 $74,625 6,385 169
2012*** $55,041 3,972 318 $84,858 6,893 197
Note: All income adjusted to 2012 dollars. Only respondents with non-zero income reported.
*** Denotes a statistically significant difference in means at p<01.
Sampling weights and adjustments for stratified cluster sampling applied.
Foregone wages and indirect costs. A key issue for this study is the cost borne
by a near completer who returns to school in the form of foregone wages. Having likely
worked for more years than traditional-aged students, they may earn higher wages and
could thus face substantially higher foregone wages if they are not able to work and
complete school at the same time. Table 11 shows mean earnings for near completers
compared to near completers who are enrolled in college and near completers who
graduated at three points in time. Enrollment is a self-reported measure and could include
both full- and part-time students. These data show that near completers who are enrolled
have substantially lower earnings (with statistical significance) in two of the three years
Table 11: Income for near completers, by enrollment and graduation status.
Near Completers, Not Enrolled Near Completers, Enrolled Near Completers, Finished Degrees
Year Mean Linearized N Mean Linearized N Mean Linearized N
Income Std. Err. Income Std. Err. Income Std. Err.
1992 $38,054 1,864 338 $27,517 2,124 76 $49,914 4,821 139
2002 $51,537 3,007 351 $43,551 10,153 30 $74,625 6,385 169
2012 $56,241 4,111 299 $26,883 5,554 19 $84,858 6,893 197
Note: All income adjusted to 2012 dollars. Only respondents with non-zero income reported.
Statistically significant difference in means at p<01 for non-enrollees and enrollees only.
Sampling weights and adjustments for stratified cluster sampling applied.


71
compared to near completers who are not enrolled. These results are suggestive that there
may be some indirect costs associated with returning to finish a degree, even for students
who are part-time. The number of enrolled students in a given year (ranging from 76
down to 19 in 2012) are relatively small, so caution is warranted before reaching
conclusions.
The Ashenfelter Dip. The panel nature of the dataset makes it possible to control
for the potential existence of the Ashenfelter Dip by examining incomes in the survey
round prior to reenrollment. For survey rounds prior to 1994, this represents the year
prior to reenrollment. For survey rounds after this period, reported income is for two
years prior to reenrollment due to the biennial nature of the survey from 1994 through
2012. The descriptive data presented in Table 12 compares incomes of near completers
with near completers who enroll in the following survey round.
Table 12: Pre-enrollment income for returning near completers
Attainment Status Mean Income Linearized Std. Err. Observations
Near Completers3 $40,510 583 7230
Pre-enrollment3 $32,743 2,191 350
Note: All income adjusted to 2012 dollars. Only respondents with non-zero income reported.
Statistically significant difference in means at p<01 for near completers and preenrollees.
Sampling weights and adjustments for stratified cluster sampling applied.
Instead of only examining individual years, I consider average salaries for all
observations of near completers and all observations where the individual is in the pre-
enrollment category. The difference shows strong statistical significance, and suggests
that there may be a decline in income prior to reenrollment that could bias estimations if
it is not accounted for in the final model. 13
13 Although this presents potential measurement issues, sensitivity testing of the regression models
presented later shows no difference if the sample is restricted to 1994 so that the income variable is truly
the year before enrollment, orthe full sample, with biennial responses from 1994-2012.


72
Direct costs of college attendance. As described in greater detail in the following
chapter, accounting for the direct costs borne by individuals returning to complete
degrees is necessary for estimating the actual rate of return. As opposed to foregone
wages, where near completers are likely to sacrifice more money than traditional
students, direct costs as a percentage of their earnings are likely to be lower than those
borne by traditional students. With the available data, however, it is not possible to
determine the actual tuition paid by any individual near completer returning to school.
This presents difficulties for estimating an actual rate of return. Instead, as described in
greater detail in the section on methodology, I present a range of scenarios with different
levels of direct cost as a percentage of a near completers earnings. In developing the
range of direct costs to use, I consider differences in direct costs by sector. Table 13
shows average tuition rates by management type over the course of the survey. While this
is only a rough estimate of what an individual might have paid for attending additional
postsecondary education, it does, in conjunction with estimates of foregone wages,
provide a basis for approximating actual rates of return, as described in greater detail in
the following chapter. One important observation from this table is that tuition prices
have increased over time, even after adjusting for inflation. This likely means that
estimates of the return to degree completion will depend to some extent on the year in
which an individual completes his or her degree. If the return to degree completion
remains constant, those who completed degrees later will bear higher costs and thus
likely receive a smaller economic return relative to those who completed earlier.
Increasing returns to degree completion over time may offset this tuition growth,
however (Athreya & Eberly, 2015).


73
Table 13: Average tuition and fees by school management type, 1980-
2012
Year Public Tuition Private Tuition*
Non-profit For-profit
1980 $2,111 $9,501
1981 $2,199 $9,947
1982 $2,391 $10,292
1983 $2,566 $10,845
1984 $2,642 $11,435
1985 $2,755 $12,106
1986 $2,892 $12,921
1987 $3,020 $13,728
1988 $3,091 $14,010
1989 $3,189 $14,601
1990 $3,208 $14,907
1991 $3,486 $15,509
1992 $3,751 $15,873
1993 $3,948 $16,453
1994 $4,056 $16,810
1996 $4,278 $17,898
1998 $4,465 $18,571
2000 $4,550 $20,270 $13,247
2002 $5,056 $21,692 $14,133
2004 $5,967 $23,147 $15,475
2006 $6,316 $24,348 $15,916
2008 $6,691 $25,975 $15,169
2010 $7,345 $27,156 $14,670
2012** $7,904 $28,533 $14,955
All figures presented in 2012 dollars.
*NCES did not calculate private tuition separately for private non-
profit and private for-profit colleges and universities until 1999.
**2012 Figures not available. Average tuition calculated by adding the
averaging change in tuition from 2000-2010.
Source: National Center for Education Statistics (2013).
Public vs. private graduates. Finally, descriptive data on near completers who
finished at public colleges and universities are compared to those who finished at non-
profit and for-profit colleges and universities in Table 14. The sector of college or
university from which an individual graduated is derived from the NLSY restricted use


74
Table 14: Descriptive data for near completers who finish a degree, by sector
Variables Public .. Linearized Mean r- Std. Err. Private, non-profit .. Linearized Mean r- Std. Err. For-profit .. Linearized Mean r- Std. Err.
Mean Income (2012) $91,264 9,401 $75,816 13,730 $70,001 15,644
Yrs. since graduation (2012)a 12.42 0.51 11.30 1.02 6.58 1.74
Mean Income (1 yr. post grad.)bc $28,712 29,916 $35,709 29,524 $49,574 18,708
Age (at graduation)bd 29.89 6.75 32.15 8.65 42.69 8.15
Magazines (1979) 0.83 0.03 0.81 0.05 0.67 0.14
Newspapers (1979) 0.89 0.02 0.94 0.02 0.78 0.15
Library card (1979) 0.89 0.02 0.89 0.04 0.69 0.16
Mothers high grade 13.01 0.25 13.32 0.33 13.17 1.19
Fathers high grade 13.70 0.34 13.90 0.47 13.06 1.79
Family below poverty line (1979) 0.06 0.02 0.07 0.03 0.03 0.03
ASVAB Score 75.16 1.66 78.04 2.46 70.14 7.15
N 197 75 13
Note: All income figures adjusted to 2012 dollars. Only respondents with non-zero income reported.
Descriptive data on gender and racial/ethnic background suppressed due to small cell sizes.
a Years since graduation shows a statistically significant difference between public and for-profit at
p<.01 and a statistically significant difference between private and for-profit at p<05.
b Due to these events occurring in different years, unweighted means are reported for these two
variables, along with traditional standard errors.
c Post-graduation income shows a statistically significant difference in means for public and for profit
at p<10. Receipt of newspapers shows statistically significant
d Ages show statistically significant differences between for-profit and both other categories at
p<.01. Differences in age between public and private are statistically significant at p<10.
Sampling weights and adjustments for stratified cluster sampling applied.
e Receipt of newspapers shows statistically significant differences at p<10 between public and non-
profit schools.
dataset, following Monks (2000). The survey includes the Federal Interagency
Committee on Education (FICE) codes to identify the school. Sector is then determined
by the Carnegie classifications. The number of individuals who graduated from for-profit
schools is quite small, perhaps due to the fact that for-profit institutions have only grown
significantly in the last decade and the number of near completers who finished degrees
has declined in recent years. While the number of near completers who finished degrees


75
at for profit schools represents only 6 percent of the total number of finishers, for profit
graduates account for 15 percent of those who have graduated since 2000.
The descriptive data show that individuals who graduate from for-profit schools
may have higher incomes immediately after graduating (although the difference is not
significant), while public school graduates have higher earnings in the final year of the
survey. The differences in earnings in the final year of the survey are statistically
significant between public and non-profit graduates only. There is a statistically
significant difference between means for the age when individuals graduate, with for-
profit graduates tending to be older on average when they finish than graduates from
other schools, which suggests that they are finishing their degrees much later in their
careers. The only other variable with a statistically significant difference is the receipt of
newspapers in the family home growing up, with non-profit school graduates more likely
to receive them than public school graduates. Likely due to the limited number of
individuals who graduated from for-profit schools, none of the other variables show
statistically significant differences.
Data Limitations. As discussed above, non-random attrition is a concern,
although previous research suggests that the level of attrition here should not overly bias
results. There are other limitations to the data that should be acknowledged. The surveys
measure of racial/ethnic background is problematic with only three categories, and this
may limit the generalizability for findings based on these sub-groups. With these data,
however, there are no alternative measures. As will be discussed in more detail in the
chapter on methodology, I combine African-Americans and Hispanics into a single
category to improve the size of the subsamples. While this does allow for analysis of


76
important differences between predominantly White individuals (although technically,
individuals of other racial/ethnic backgrounds could be included), it may mask important
differences between Hispanics and African-Americans. Additionally, key variables such
as education level and income are self-reported, which could be a concern (Ashenfelter &
Krueger, 1994). Should measurement error for these variables be random, it would not
bias estimates but only increase standard errors (Wooldridge, 2006). However, if
individuals systematically overstate or understate their earnings, education level, or other
key variables, it could bias estimates (Bound & Krueger, 1991; Wooldridge, 2006). In the
particular models presented in the following chapter, where income is a dependent
variable, bias could occur if the tendency to misreport ones income is correlated with
any of the independent variables used (Bound & Krueger, 1991). A study comparing both
self-reported income for a longitudinal survey and income reported to the Social Security
Administration by an individuals employer finds that although there are statistically
significant differences between the two, they are substantively small and should not
overly bias estimates (Bound & Krueger, 1991).
Self-reported education is also subject to non-random error. Ashenfelter and
Krueger (1994) find that about ten percent of the variance in self-reported schooling
levels (when years of schooling are reported) is due to error. Other research suggests that
individuals are much more accurate in reporting degree levels attained than in reporting
years of schooling (Kane, Rouse, & Straiger, 1999). While this does raise some concern
for the estimates I present below, my estimations do not attempt to examine the marginal
return to a year of postsecondary education, which minimize bias. Further, the decision
by NLSY administrators to switch to computer assisted interviews has been shown to


77
reduce errors and improve accuracy in self-reported data (Baker, Bradbum, & Johnson,
1995; Tourangeau& Smith, 1996).
The data used in this analysis only consider the type of school from which a near
completer graduates. Many returning students may reenroll multiple times before starting
the enrollment spell that leads to degree completion. More granular analysis of this
process would likely benefit policymakers; however, it is beyond the scope of this study.
Further the analysis of outcomes by sector is somewhat limited by the uneven distribution
of graduates in each sector, especially the small number of graduates from for-profit
schools.
Data: Conclusion
In spite of these potential limitations, the NLSY dataset offers substantial depth of
information and its continuity since 1979 provides a rich longitudinal cohort. Descriptive
data and initial analyses are suggestive that near completers who finish degrees may
receive substantial wage premiums; however, it could also be that they forego wages in
pursuit of these degrees. These initial analyses suggest that models will have to control
not just for characteristics like race, gender, and family background, but also an
individuals income prior to and during college enrollment. Initial analyses of the
outcomes for graduates of different sectors show suggestive differences in incomes
between sectors, as well as differences in the age at which individuals graduate, but
further investigation is warranted.


78
CHAPTER IV
METHODOLOGY
To address the research questions posed at the outset of this dissertation, I employ
two series of models. The first series is designed to estimate the income benefit from
completing a degree, including estimates for the income benefit from graduating from
public, private, and for-profit colleges and universities. While these models provide
estimates of the wage premium for completing a degree, this does not represent the actual
economic return, except under a set of assumptions that are unrealistic for near
completers. The following section of this chapter explains how I use estimates of the
wage premium, along with other data, to determine the wage, tuition, and work-life
conditions under which a near completer who finishes a degree is likely to earn a positive
economic return. This chapter concludes with a section describing the event history
analysis model used to determine how certain observable characteristics affect the
likelihood that an individual will return to finish his or her degree.
Individual-Level Fixed Effects
As noted in the review of literature above, ability bias is one of the key sources of
endogeneity in research on the economic return to education. The concern is that
unobservable characteristics, such as ability or motivation, may cause near completers to
return to finish a degree, while also causing them to earn higher wages independent of
their education levels. The models and methodological approach here seek to control, to
the greatest extent possible, these and other sources of bias that could impact findings.
The models in this series attempt to estimate changes in income that are due to an
individual near completer returning to finish his or her degree. Although the dataset


79
offers a rich collection of control variables for each individuals background and ability,
there is still concern about omitted variables and the endogeneity of education,
particularly if the ASVAB score does not fully capture innate ability. I take advantage of
the longitudinal nature of the data to employ an individual-level fixed effects estimation,
which compares each individuals pre- and post-treatment earnings, and in doing so,
controls for all time-invariant characteristics of individuals that could bias results
(Allison, 2009; Angrist & Pischke, 2009; Cellini & Chaudhary, 2014; Jepsen, Troske, &
Coomes, 2014; Wooldridge, 2006). Thus, assuming that an individuals innate ability is
time-invariant, this approach will eliminate the potential for ability bias. Additionally,
this approach controls for observable characteristics that do not vary over the time period
in question, such as gender, racial/ethnic background, parental education, high school
achievement, and the characteristics of the individuals family during his or her youth.
The model is specified as follows:
Yit = Hi + Pidit + p2xu +o-i + t, + elt
where Y is the logged annual earnings (adjusted for inflation) for individual i in year t.
The key education variable is d, which is a series of dichotomous variables for four
potential education levels: near completer; near completer just prior to enrollment; near
completer, enrolled; and near completer, finished degrees. The education attainment
variables are mutually exclusive in any given year. In the analyses that follow near
completer is omitted and that category serves as the comparison to the other categories.
The additional categories help eliminate two potential sources of bias. The
Ashenfelter Dip may lead to biased estimates if incomes decline prior to reenrollment,
and then naturally recovers. Similarly, if there are substantial foregone wages while near


80
completers are enrolled in school and working to finish a degree, estimates for the degree
completion coefficient would be biased as these individuals finish school and no longer
lose earnings because they are enrolled.
Additionally, the model includes x, which is a vector of time-variant
characteristics of individual i in year i, including age, experience, weeks out of the labor
force, weeks unemployed, number of children, and health status. Year dummy variables
are included in r, while u is the individual intercept, which includes the time-invariant
observable and unobservable characteristics of individual z, such as gender, racial/ethnic
background, innate ability, ASVAB scores, parental education, family socio-economic
status, high school academic success, and innate ability, e is the individual error term at
each year t; a represents the combined error of all time-invariant variables with relation
to y (Allison, 2009).
Following standard practice in return to education literature, a quadratic term for
experience here a derived variable counting each year an individual works at least 26
weeks is also included (Becker, 1993; Mincer, 1974). However, some research shows
that limiting the specification to a quadratic experience term introduces significant bias
into the model, as the error term is correlated with experience (Murphy & Welch, 1990).
Including only the quadratic term may understate income growth early in an individuals
career, and overstate the late career decline (Murphy & Welch, 1990). To correct for this
potential bias, I include higher order experience terms.
Because the individual-level fixed effects approach absorbs several variables that
are of interest into the term that includes all time-invariant characteristics, I repeat the
model using interaction terms for racial/ethnic background and gender. This approach


81
shows whether there are differences in the impact of certain characteristics depending on
an individuals sub-group membership and whether sector of graduation impacts the
wage premium an individual receives.
The individual-level fixed effects approach essentially captures the treatment
effect of graduation by comparing pre- and post-degree earnings. This approach is not
suitable for many analyses of the economic returns to education because for traditional
students that proceed directly through the education pipeline, their actual earnings prior
to earning a degree are likely quite low due to the types of employment available to full-
time students. However, with near completers, who have opportunities to work full-time
before treatment, this approach is appropriate and can control for major sources of
omitted variable bias (Cellini & Chaudhary, 2014; Jepsen, Troske, and Coomes, 2014).
To determine whether racial/ethnic background, gender, initial SES, and the
sector of the school from which the individual graduates affects his or her earnings, I
repeat the model as follows:
Yit = p, + PiPu + P&u +p30-i + t, + elt
The only change from the initial model is replacing the education status term with p
which is a series of interaction terms that reflect the same education categories as above,
combined with these important time-invariant characteristics. The interaction is applied
to time periods prior to enrollment, enrollment itself, and post-graduation. This will help
determine whether there are differences in wage premiums and foregone wages by
race/ethnicity, gender, initial SES, and sector.
Survey weights and regression. The question of whether to apply survey
weights provided by NLSY to the data used in the regression analyses described above is


82
complicated. Weights were applied to derive the univariate descriptive statistics
presented above, which is necessary to present data that are representative of the
population (Solon, Haider, & Wooldridge, 2015). Given that NLSY oversamples several
population subgroups, it could be argued that weighting the data for the regression
analysis would also be an appropriate course of action to provide more accurate estimates
of the effect of degree completion on earnings.
However, methodological research suggests that this may not be the best approach
given the model proposed above. While applying survey weights to the dataset makes
intuitive sense, several econometricians show that this can lead to inefficient estimates
and errant conclusions. Solon, Haider, and Wooldridge (2015) identify three typical
justifications for weighting data: the first is correcting for heteroskedasticity due to
differences in group sample size when the unit of analysis is not an individual; the second
involves trying to identify a population effect that is generalizable and representative; and
the third is to correct for endogeneity of the sample, whereby the dependent variable is a
function of the different criteria used to establish subpopulations in the sampling scheme.
Winship and Radbill (1994) similarly identify errors with using sample weights in OLS
regressions and argue that it should only be done when the sampling criteria are a
function of the dependent variable (as in the third reason cited above). In those cases, if at
all possible, they argue that the model should be respecified rather than use sampling
weights, which can result in inefficient and biased estimates.
Of these three reasons for using sampling weights, heteroskedasticity due to
different group sizes is not relevant because I use individuals as the unit of analysis.14
14 As discussed in the results, post-regression diagnostics do show that other types of heteroskedasticity are
present, but these are accounted for with clustered robust standard errors.


83
Attempting to identify an average population effect could also be justification for
weighting, particularly when heterogeneous effects may be present (Solon, Haider, &
Wooldridge, 2015). However, the preferred approach (and the one that I take here) is to
model the heterogeneous effects and report the differences rather than attempt to
determine an average across the population (Solon, Haider, & Wooldridge, 2015).
Finally, the endogeneity of the outcome variable (income) to the sampling criteria is
accounted for by the fixed effects model, which controls for all of those criteria (which
are by their nature time invariant over the course of the survey). Thus, sample weights are
not employed in the main regression analyses. Because there are still differing viewpoints
on using weights in an analysis such as this, I present a weighted regression analysis in
the appendix.15
Calculating Rate of Return
The preceding models will estimate the change in wages due to a near completer
returning to finish his or her baccalaureate degree. Under a strict set of assumptions, such
models estimate an individuals rate of return: 1) there are no costs for schooling (direct
or indirect); 2) schooling does not affect length of work life; 3) there are no taxes on
additional income earned; and 4) additional schooling does not affect the impact of
experience (or other variables) on earnings (Bjorklund & Kjellstrom, 2000; Heckman,
Lochner, & Todd, 2008). Because these assumptions do not hold up in reality
(particularly for near completers), calculating the actual return on investment requires
additional calculations.
15 As is discussed in detail in the appendix, the weighted regression shows a slightly smaller effect size for
degree completion, though it maintains its strong statistical significance. It is not possible to test whether
there is a statistically significant difference between the two effect sizes.


84
Most earnings functions based on Mincers work do not account for individuals
pursuing additional education later in life, when they may have fewer years in which to
recoup costs. Students who pursue higher education directly from high school have a long
working life which allows them to earn back the investment (both in direct and indirect
costs) for college attendance. For those deciding whether to enroll later in life, it could be
that shorter potential earning windows make it difficult to realize a positive return on
their investment. Approximations of a return to degree completion should project out
over time to account for varying time remaining in the workforce.
For the purposes of this study, taxes are ignored as research suggests the
progressive tax rate in the United States can reduce the income premium for degree
completion, but its overall impact is limited (Heckman, Lochner, & Todd, 2008). The
progressive tax rates that are a feature of the United States tax code mean that the higher
wages earned by near completers who finish degrees are taxed at higher rates, reducing
its overall benefit. However, Heckman, Lochner, & Todd (2008) find that accounting for
the tax code reduces wage premiums by less than one percentage point, so for the
purposes of this study, taxes are ignored.
I begin the analysis of whether near completers who finish a degree receive a
positive economic return with the broader model originally proposed by Becker (1993)
and adapted from Heckman, Lochner, & Todd (2008):
_ 2jt=T(i _|_ r)t Z,t=o _|_ r)t
yTo Y0,t + Ct
Lt=o + ry
where Y} represents the earnings for an individual with a college degree at time t, while
Y0 represents the earnings for a near completer at time t. r represents a discount rate,


85
while r represents the number of years an individual takes to complete a degree. The first
term in the numerator sums discounted earnings from t= rto /'/ which is the retirement
age for an individual with a college degree. The second term in the numerator and the
denominator sums discounted earnings from t=0 to To, which is the retirement age for a
near completer. The direct costs for attending college are included in the term Ct where C
represents costs for education in year t, so if the individual is not enrolled in that year,
this term will be zero. As noted earlier, this model only accounts for foregone wages by
assuming that the individual does not work at all for r years, by beginning the summation
in the year r rather than year 0.
I modify this model by first assuming that individuals will work until the same
age. Second, I explicitly include foregone wages as a separate term (Ft). The modified
equation is as follows:
vr ^i,t Y0it Ft Ct
Zit=o
R =
(1 + r)t
yTo Yp,t + Ft + Q
Lt=o (l + r)t
The equation can be further simplified by substituting the wage premium, written as a
percentage of Yo, in place of the Y o, term and carrying out a similar procedure for the Ft
term. This information can be supplied through the estimates of wage premiums and
foregone wages from the fixed effects model. The remaining unknown variables will be
the direct cost of attendance and the length of time remaining in the workforce.
Because there are not data on the actual tuition costs paid by individual near
completers, it is impossible to specify each individuals direct costs. Instead, I present a
range of scenarios where tuition is calculated as a percentage of an average near
completers earnings. This helps simplify the above equations so that the coefficient for


86
wage premiums, the direct costs, and the coefficient for foregone wages are all presented
as a percentage of earnings. Given the numerous potential avenues for discounted tuition
(including grants, scholarships, employer assistance, and military benefits, as well as
part-time enrollment) I use a range of different tuition figures to calculate these potential
direct costs, using 25 percent, 50 percent, 75 percent, and 100 percent of each sectors
average reported tuition at different points in time and calculate that tuition figure as a
percentage of average earnings.
Rather than producing a definitive rate of return for individuals attending different
types of schools, this approach will produce an estimated rate of return dependent on
assumptions about the level of direct costs borne by an individual. While this approach is
imperfect in some ways, it is superior to ignoring direct costs altogether.
The final unknown variables are the length of time remaining in the workforce
and the number of years an individual is enrolled. A near completer who finishes a degree
and only works for one year following graduation would likely not see a positive
economic return (unless the wage premium was extremely large). It may take many years
of employment for the wage premium to cover the costs borne by the individual,
particularly when using an appropriate discount rate. Additionally, a near completer who
takes several years to graduate and bears substantial foregone wages during these years,
will also likely earn a smaller economic return, although a part-time student may forego
fewer wages. These different contingencies are accounted for in the different scenarios
presented in the results chapter below.
Plotting the results produces a curve with a negative initial return (when a near
completer is paying direct costs and sacrificing wages) that increases over time. Whether


87
the plot becomes positive, and if so, how long it takes and how high it gets will provide
clarity on the economic return for near completers who finish degrees. Although such
plots are not common in other literature on the returns to education, they make sense for
the population under consideration here. Similar plots are used in investing when there
may be negative returns early in the investment period, followed by a period of
increasing returns.16 A hypothetical plot is included below.
0 5 10 15 20 25 30 35 40 45
Years
Figure 7: Plot of hypothetical rate of return for near completer finishing a degree
In this example, year 0 represents the decision to return to finish a degree. This individual
remains enrolled for two years, at which point he or she has received a substantially
negative return of about 25 percent. Following an assumed graduation at year two, the
return begins to climb as the individual earns a wage premium for his or her degree,
crossing the break-even threshold at about year 5. The return levels off at just over 10
percent. In the results section, I present a series of similar plots approximating the return
to degree completion under a series of different tuition scenarios.
16 See for example Understanding the J-Curve: A Primer on Interim Performance of Private Equity
Investments: http://www.goldmansachs.com/gsam/pdfs/USTPD/education/understanding_J_Curve.pdf


88
Event History Analysis
To evaluate hypothesis 3, which focuses on whether an individuals
characteristics affect the probability that he or she will return to finish a degree, I employ
an event history model. A standard probability analysis using a probit or logistic
regression model is not appropriate for these longitudinal data because of two
characteristics: censorship and time-varying explanatory variables (Allison, 1982;
Mills, 2011). Censorship refers to the fact that even though a near completer did not
graduate during the survey period, there is still a possibility that he or she might graduate
during subsequent periods (Allison, 1982; Mills, 2011). Further, in a probit or logistic
regression model, an individual who becomes a near completer late in the survey period
but never finishes a degree would have as much impact on the estimations as someone
who became a near completer early in the survey period and remained so for 25 years.
Additionally, these approaches cannot account for changes in explanatory variables over
time (Allison, 1982; Mills, 2011). These shortcomings are all relevant for the analyses in
question.
Essentially, the event history analysis employed here calculates how different
variables affect the likelihood that the event of interest will occur (in this case, the event
is the decision to return to school for an enrollment spell that results in degree
completion.) Owing to its origins in medical and biological research, this methodology
is also known as survival analysis. The occurrence of the event of interest is known as a
failure (which in the case of medical research often means the death of a subject) and 17
17 Due to the fact that many variables of interest particularly income are affected by enrollment status, it
would not be appropriate here to model graduation as the event of interest.


89
18
the time until that event occurs is known as the survival time. In this case, a longer
survival time means that an individual who has become a near completer is taking longer
to return to finish his or her degree.
There are multiple options for selecting a model to estimate the impact of
different variables on the likelihood of graduation, including semi-parametric models
such as the Cox proportional hazards model and parametric models. The latter set of
models can be carried out in two different forms: accelerated failure time (AFT), which
estimates the effect of covariates on increasing or decreasing survival times, and
proportional hazards models, which estimate how different covariates affect the hazard
rate. This is the likelihood of failure occurring in a time period given that it has not
occurred prior to that (Bradburn, Clark, Love, & Altman, 2003; Jenkins 2008; Mills,
2011; Taniguichi & Kaufman, 2005). Additionally, within the family of parametric
models, there are numerous choices for specifying the distribution of the hazard over
time, including exponential, Weibull, Gompertz, log-logistic, log-normal, and gamma
models (Bradburn, Clark, Love, & Altman, 2003; Jenkins 2008; Mills, 2011). Choosing
between these approaches depends on theoretical insights into the shape of the hazard
rate, supported by a series of post-estimation diagnostics (Box-Steffensmeier & Jones,
2004; Bradburn, et al., 2003; Jenkins, 2008).
The Cox proportional hazards model does not require this specification between
distributions and is consequently a popular approach in the literature (Bradburn, et al.,
2003; Mills, 2011; Orbe, Ferreira, & Nunez-Anton, 2001). Although I employed this
model initially, as discussed in greater detail below, post-estimation diagnostics showed 18
18 Although the nomenclature that considers returning to finish a degree a failure is unfortunate, I use it
here because it is relevant to the choice of an Accelerated Failure Time (AFT) model.


Full Text

PAGE 1

THE RETURN ON RETURNING: THE ECONO MIC BENEFIT OF BACCALAUREATE DEGREE COMPLETION AF TER STOPPING OUT by PATRICK DAVID LANE A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Doctor of Philosophy Public Affairs 2015

PAGE 2

ii This thesis for the Doctor of Philosophy degree by Patrick David Lane has been approved for the Public Affairs Program by Paul Teske, Advisor Todd Ely Chair Kelly Hupfeld Susan Clarke Date: 11/15/2015

PAGE 3

iii Lane, Patrick David (Ph.D., Public Affairs) The Return on Returning: The Economic Benefit of Baccalaureate Degree Completion after Stopping Out Thesis Directed by Professor Paul Teske ABSTRACT An emerging strategy in higher education and workforce development policy circles aims to raise local, state, and national degree attainment rates by target ing those who left postsecondary education after earning significant college credits but without completing a degree. This dissertation examine s some of the assumptions behind these programs test ing receive a positive economic return compared to those who do not return to finish a degree. Additionally, this research examines whether their outcomes are impacted by the sector (either public private non profit, or private for profit ) of the college or university at which they complete their degr ee. Finally this study examines whet her individual characteristics affect the likelihood that an individual who has stopped out of college will return to complete a degree. Overall, I find that the economic return varies across racial/ethnic background an d that not all subgroups earn a positive return from finishing a degree, but returns do not differ by sector. Finally, I find that many of the factors generally associated with increased educational attainment do not appear to have a relationship with the likelihood of finishing a degree The form and content of this abstract are approved. I recommend its publication. Approved: Professor Paul Teske

PAGE 4

iv DEDICATION For Sara, without whom this simply would not have been possible. And for Cora, without whom this still would have been possible (and been possible quite a bit sooner) but whose absence would have made life much less enjoyable.

PAGE 5

v ACKNOWLEDGEMENTS This dissertation owes its existence to the guidance, support, and encour agement of numerous individu als. Dr. Paul Teske guided this work and provided well balanced leadership, advice, and direction throughout the process as my advisor. Dr. Todd Ely also provided thoughtful feedback and patient explanations of countless quantit ative concepts from the very inception of this research both in his faculty role and as the chair of the dissertation committee. Drs. Kelly Hupfeld and Susan Clarke helped fine tune the research and pushed me to consider several additional angles that have resulted in a much stronger overall research effort. This work would also not have been possible without the support and advice of colleagues at the Western Interstate Commission for Hig her Education (WICHE), where I have been employed throughout this pr ocess Longanecker, has established an organizational culture in which all staff are encouraged to pursue additional education while Dr. Demar e Michelau was extraordinarily supportive throughout the project and an invaluable source of advice and guidance. Dr. Brian Prescott has also contributed through encouragement and support, including in helping to obtain the data necessary to complete the research. Finally, this work owes a debt of gratitude to Dr. Peter deLeon. Though h e may not be aware, the initial meeting I had with him six years ago convinced me that pursuing

PAGE 6

vi TABLE OF CONTENTS CHAPTER I: INTRODUCTION ................................ ................................ ................................ ............ 1 Significance of the Study in Practice ................................ ................................ .............. 3 Outcomes at Public and Private Colleges and Universities ................................ ............ 8 Practical Implications Conclusion ................................ ................................ .............. 11 Research Questions ................................ ................................ ................................ ....... 12 II: L ITERATURE AND HYPOTHESES ................................ ................................ .......... 14 The Economic Return to Degree Completion ................................ ............................... 15 Distinctions between Public and Private Organizations ................................ ............... 35 Literature Review: Conclusion ................................ ................................ ...................... 42 Hypotheses ................................ ................................ ................................ .................... 43 Contributions to the Field ................................ ................................ .............................. 46 III: D ATA AND MEASUREMENTS ................................ ................................ ............... 53 Data Source ................................ ................................ ................................ ................... 53 Data: Conclusion ................................ ................................ ................................ ........... 77 IV: M ETHODOLOGY ................................ ................................ ................................ ...... 78 Individual Level Fixed Effects ................................ ................................ ...................... 78 Calculating Rate of Return ................................ ................................ ............................ 83 Event History Analysis ................................ ................................ ................................ .. 88 Methodology: Conclusion ................................ ................................ ............................. 91 V: R ESULTS ................................ ................................ ................................ ..................... 94 The Wage Premium for Degree Completion ................................ ................................ 94

PAGE 7

vii Returns to Degree Completion ................................ ................................ .................... 104 Why Do Near Completers Return? ................................ ................................ ............. 117 VI: D ISCUSSION & CONCLUSIONS ................................ ................................ ........... 126 Economic Returns to Degree Completion ................................ ................................ ... 127 Why Do Near Completers Return? ................................ ................................ ............. 134 Conclusions ................................ ................................ ................................ ................. 137 REFERENCES ................................ ................................ ................................ ................ 143 APPENDIX ................................ ................................ ................................ ...................... 157

PAGE 8

viii LIST OF TABLES TABLE 1: College enrollment by age group and sector, 2011 ................................ ...................... 11 2: Returns to degree completion ................................ ................................ ....................... 19 3: Estimations of the sheepskin effect ................................ ................................ .............. 22 4: Research questions and hypotheses ................................ ................................ .............. 45 5: College and state definitions for degree completion programs ................................ .... 55 6: Descriptive statistics for those who attained near completer status .............................. 58 7: Changes in descriptive data (unweighted) 1979 to 2012 ................................ ........... 60 8: Weighted descriptive data, 1979 and 2012 ................................ ................................ ... 60 9: Demographic data by educational attainment ................................ ............................... 64 10: Income by educational attainment 1992, 2002, & 2012 ................................ .......... 70 11: Income for near completers, by enrollment and graduation status. ............................ 70 12: Pre enrollment income for retu rning near completers ................................ ................ 71 13: Average tuition and fees by school management type, 1980 2012 ............................ 73 14: Descriptive data for near completers who finish a degree, by sector ......................... 74 15: Methodological approaches used to test hypotheses ................................ .................. 92 16: Effects of education status on earnings ................................ ................................ ....... 97 17: Differential wage premiums by gender ................................ ................................ .... 100 18: Differential effects by racial/ethnic background ................................ ...................... 101 19: Differential effects by 1979 poverty status ................................ ............................... 102 20: Income differences by sector of graduation ................................ .............................. 103 21: Approximated rates of return for baccalaureate degree completion ......................... 117

PAGE 9

ix 22: Event history analysis of degree compl etion ................................ ............................ 119 23: Event history analysis with gender interaction ter ms ................................ ............... 121 24: Event history analysis with racial/ethnic interaction terms ................................ ...... 122 25: Event history analysis with poverty in teraction terms ................................ .............. 123 26: Definitions of near completer ................................ ................................ ................... 157 27: Regressions with different definitions of near completer ................................ ......... 158 28: Differential effects by gender ................................ ................................ ................... 160 29: Differential effects by race/ethnicity ................................ ................................ ........ 161 30: Differential effects by familial poverty status ................................ .......................... 162 31: Income differences by sector of graduation ................................ .............................. 163 32: Event history analysis u sing different near completer definitions ............................ 165 33: Event history analysis with gender interaction terms ................................ ............... 166 34: Event history analysis with racial/ethnic interaction terms ................................ ...... 167 35: Event history analysis sensitivity tests with poverty status interactions .................. 168 36: Categorical years since graduation ................................ ................................ ........... 169 37: Continuous years since graduation ................................ ................................ ........... 170 38: Comparison of weighted and non weighted regressions ................................ .......... 172

PAGE 10

x LIST OF FIGURES FI GURE 1: Near completers and near completers finishing degrees over time ............................... 63 2: Near completers and near completers finishing degrees ove r time, accounting for attrition ................................ ................................ ................................ ................... 63 3: Survival estimates for near completers ................................ ................................ .......... 66 4: Survival estimates for near completers by gender ................................ ......................... 67 5: Near completer survival rates by racial/ethnic group ................................ .................... 68 6: Near completer survival rates by familial poverty status ................................ .............. 69 7: Plot of hypothetical rate of return for near completer finishing a degree ...................... 87 8: Public tuition as a perce ntage of mean wages of near completers ............................... 106 9: Non profit tuition as a percentage of mean wages of near completers ........................ 106 10: For profit tuition as a percentage of mean wages of near completers ....................... 106 11: Public tuition as a percentage of earnings of African Americans/Hispanics ............ 107 12: Return for baccalaureate degree completion public tuition rates ........................... 110 13: Return for baccalaureate degree completion non profit tuition rates ..................... 111 14: Return for baccalaureate degree completion public tuition rates ........................... 112 15: Return for baccalaureate degree completion non profit tuition rates ..................... 113 16: Return for baccalaureate degree completion for profit tuition rates ....................... 113 17: African American/Hispanic return for baccalaureate degree completion public tuition rates ................................ ................................ ................................ ........... 115 18: African American/Hispanic return for baccalaureate degree completion non profit tuition rates ................................ ................................ ................................ ........... 115 19: African American/Hispanic return for baccalaureate degree completion for profit tuition rates ................................ ................................ ................................ ........... 116 20: African American/Hispanic rate of return with varied indirect co sts ........................ 171

PAGE 11

xi

PAGE 12

1 CHAPTER I INTRODUCTION Within higher education and workforce develo pment policy circles, there is a heavy focus on increasing the percentage of the United States population that has a postsecondary degree or certificate These efforts include a wide range of public sector program s and policie s designed to increase th e rates at which individuals choose to attend college and also boost the percentage of those who finish credentials once they enter. There is strong evidence that increased degree attainment rates have soc ietal benefits as well as financial returns to the individuals completing degrees (see for example, Hout, 2012). Within a wide range of po licy efforts to increase postsecondary credential attainment, many state governments a re including a special focus on those individuals who earn ed significant college credits before leaving, working on the intuitive assumption that they will be able to complete degrees and certificates more quickly and at a lower cost (see for example Minnesota State Colleges and Universities, 2015; Oklahoma State Regents for Higher Education, n.d.) Underlying this work is an assumption that these individuals who finish degrees will earn substantial financial benefits; however, little research has examined the question Further, few, if any, studies examine the fa a term I use to denote those individuals who finish significant college credits but do not earn a degree will return to finish baccalaureate degrees 1 1 Although many policy efforts include a focus on encouraging those with prior college credit to return to only.

PAGE 13

2 Although the general question about whether or not there is an economic benefit to education may be settled, research on near completers is an unexplored avenue. While labor economists have reached degree is like ly to produce an individual economic return, t here has been no pr evious research looking at the outcomes of near completers who finish degrees Many of the data cited to support these programs (and to encourage nea r completers to enroll) compare incomes fo r degree holders, which is not the a ppropriate comparison. The appropriate way to analyze income gains by near completers is to compare their earnings after they graduate to near completers who had completed a similar amount of postsecondary education. In addition to addressing this key question for those returning to complete baccalaureate degrees this study builds on public affairs literature on public and private organizations by analyzing whether the economic benef its earned by near completers who finish baccalaureate degrees are dependent on whether they return to a public private non profit, or private for profit college or university Finally, in this dissertation I explore the factors that affect the likelihood that a near completer will return to finish his or her baccalaureate degree, which is of interest for education and workforce development researchers as well as policymakers and practitioners. This study is organized as follows. After this brief introduct ion to the topic, the importance of this research to the higher education and workforce development policy community is discussed, noting several important gaps in empirical know ledge that will be filled by these results. Following this introduction I pre sent formal research questions, completing Chapter 1 : Introduction Chapter 2 : Literature and Hypotheses consists of a

PAGE 14

3 review of relevant literature followed by hypotheses derived from the literatur e. Chapter 3 : Data and Measurements focuses on describing the data and measures used for the analysis. Chapter 4 : Methodology presents the methodological approaches used to test my hypotheses. Chapter 5 : Results presents the results of the empirical analyses, and Chapter 6 : Discussion and C onclusions concludes the study with a discussion of the results, their implications, and avenues for further study. Significance of the Study in Practice This research contributes to both the labor economics literature on human capital and signaling as we ll as public affairs literature on public and private organizations. Addition ally, this research inform s current policy debates on degree completion programs In this section I discuss the relevance to policymakers and practitioners by providing some back ground on degree completion and near completers as well as the programs being implemented to bring them back to college The contribution that this proposed study will make to the academic literature is discussed following the review of literature in Chapt er 2 Increa sing societal d egree a ttainment There is currently a strong push in education and workforce development policy circles to increase the percentage of the adult population in the United States that has a postsecondary degree or certificate. At the societal level, arguments for this increased degree attainment generally hinge on projections about future workforce demands and degree production rates while arguments aimed at individuals emphasize increased wages and better life outcomes The soc ietal level imperative for increasing degree attai nment rates is based on research showing that e mployer demand for individuals with college degrees will outstrip

PAGE 15

4 degr ee production by 2025 between the number of jobs that will require postsecondary education and the number of individuals with adequate qualifications (Carnevale, Smith, & Strohl, 2010). This projected gap has driven the setting of ambitious national and state goals for degree att ai nment. Lumina Foundation, which is a large, private foundation focused on postsecondary education outcomes, established a goal of having 60 percent of the adult population with a postsecondary degree or certificate by 2025 to provide the necessary educated workfo rce to meet future employment demands (Lumina Foundation, 2011). A bout 50 percent of the adult population currently has a degree or certificate (Ewert, 2013; Lumina Foundation, 2013). 2 The Obama administration established a similar goal of having the highe st percentage of 25 34 year olds with postsecondary credentials in the world by 2020 (U.S. Department of Education, 2011). Currently, the U.S. ranks 16th (White House, n.d.). Numerous states and even cities have also set aggressive goals for increasing deg ree attainment (see for example 55,000 Degrees, 2014; HCM Strategists, 2014). Increasing degree completion by adults and near c ompleters Efforts to increase degree completion by adults generally thought of as being over 25 have grown because it will not be possible to meet the goals cited above even with massive individuals that go directly from high school to college and complete degrees) ( National Center for Higher Education Management Systems and Delta Project on College Costs, 2011). Out of the broader adult population, policymakers have naturally focused on the 2 T here is some debate about this figure because current data systems do not adequately track certificates, which have a broad definition of any non degree credential.

PAGE 16

5 census category meaning that they have finished some college credits without completing a degree (U.S. Census Bureau, 2011). The logical conclusion is that these individuals would have an easier time completing a degree than th ose starting from zero credits. This broad ce nsus category includes all those who finished any college credits, meaning that individuals in this category may have only finished one class, while others may be within a few credits of a degree. The strategy of targeti ng near completers is spreading. Many states are now pursuing programs explicitly aimed at this population but t he assumptions behind these efforts are untested. Chief among these is the assumption that near completers who finish degrees receive an individual economic benefit (see for ex ample Kentucky Council on Postsecondary Education, 2008; Oklahoma State Rege nts for Higher Education, n.d.). Messages based on this assumption tend to form the foundation o f marketing and outreach efforts that encourage ne ar completers to return to postsec ondary education to finish a degree (see for example Adult College Completion Network, n.d.; Minnesota State Colleges and Universities, n.d .) This conventional wisdom tends to be based on see for example Baum, Ma, & Payea 2013 ). This comparison is inadequate for several reasons. First, as noted earlier, this category includes individuals who have earned any credit s. Those who have earned relatively few credits may have different characteristics than those who earned significant credits and would likely pull down the average wages of this group. Second, given that a ory is likely to have completed degrees

PAGE 17

6 on a more direct path than near completers, it is tenuous to assume that near completers will realize a similar wage premium from finishing their degree. Third, these comparisons do not usually examine important indi vidual characteristics that may also affect the earnings of near completers who finish degrees. A better comparison is to examine how near completers who finish degrees fare economically compared to near completers who do not finish degrees, while appropri ately controlling for individual characteristics that may also influence earnings. Beyond uncertainty about the economic benefits for individuals who return to finish degrees, t here are several additional s hortcomings in the research on which these progra ms are based There is little research on the credit distribution and demographic nearly impossible to tell how many are close to earning a degree and how many earned only a few credits before leaving postsecondary education. In spite of this uncertainty, many states are assuming that there are large enough numbers of residents close to earning degrees to quickly raise attainment levels. Further, res earch has not yet examined whether the types of degrees these near completers are likely to earn will match future workforce needs to address the degree gap. Finally, some state programs focus on the idea that near completers who finish degrees will increa se tax revenue through their higher salaries which is another unproven assumption (Abdul Alim, 2011). 3 3 s automatically leads to increased tax revenue is flawed. If an individual earns higher wages, he or she will pay higher taxes, but it could be that this individual took a job that some other person would have taken. Overall taxes collected by a government will only increase if the aggregate wages in that economy increase.

PAGE 18

7 One recent report examines those who have completed two or more years of progress toward a degree within the last 20 years (Shapiro, Dundar, Yuan, Harrell, Wild, & Ziskin, 2014). This research is informative as it compares this group, which Shapiro, et shows that the potential completers make up about 11 percent of the some college, no degree group, it does not track these individuals longitudinally to determine graduation patterns and is only able to prov ide gender among demographic characteristics (Shapiro, et al., 2014). Policies and programs focused on near completers have also g enerally not considered how such efforts impact different sub populations. A ttention elsewhere in higher education policy circ les focuses on reducing the gap in degree attainment between racial/ethnic groups and providing additional supports to low income students ( Bailey & Dynarski, 2011 ; Holzer & Dunlap, 2013) Without simple demographic data about near completers, it is not po ssible to say whether these programs could help reduce racial/ ethnic and income disparities and degree attainment or whether such programs may have different impacts on important subpopulations Further, research has not examined the characteristics of nea r completers that affect the likelihood that they will return to finish a degree. Such information would help policymakers adjust programs, target messages, and provide important supports to increase the number of near completers who return to finish degre es. While this research does not fill all of the gaps facing policymakers, it is an important step forward in providing key information about near completers by examining the economic return earned by those near completers finishing

PAGE 19

8 baccalaureate degrees a s well as the factors that affect the likelihood that near completers will return to finish such a degree Although near completers who may return to complete population is beyond the scope of this study. Outcomes at Public and Private Colleges and Universities Within h igher education research and policy circles, there has also been substantial attention paid to the differences in outcomes between public and private colleges and universities ( Cellini & Chaudhary, 2014 ; Jacobs, 2013; Monks, 2000; Schlesinger, 2010). Exami ning the outcomes for near completers who finish degrees at different types of colleges and universities will also be an important contribution to the policy conversation about degree attainment. Leaving aside, for the time being, more nuanced characteriza tions about what makes an organization public, private, or something in betw een, there are important discussions within higher education policy circles about the outcomes of students who attend public, private non profit, and private for profit colleges or universities. where high achieving, low income students tend to enroll in non selective public or private for profit college s or universities instead of attending selective, often private, non profit schools for which they qualify (Bastedo & Flaster, 2014; Hoxby & Avery, 2012; Hoxby & Turner, 2013). Interest here focuses both on the cost of attendance, which can often be minimal at private, non profit colleges and universities due to grants and scholarships offered to low income, high achieving students, and the graduation rates,

PAGE 20

9 which some research suggests would be higher for these students at the more selective schools (Hoxby & Avery, 2012; Hoxby & Turner, 2013). In another exam ple, the U.S. Senate Health, Education, Labor, and Pensions Committee has held several hearings and issued multiple reports since 2010 about the role of private, for profit colleges and universities in educating students (U.S. Senate Health, Education, Lab or, and Pensions Committee, 2012 ; U.S. Government Printing Office, 2010 ). The committee has focused on whether these private, for profit colleges and universities, which receive indirect federal funding through federal student aid programs, provide adequat e outcomes for students at a reasonable cost. One particular thread of this work is how milita ry veterans fare when entering for profit colleges and universities. These veteran students tend to be older than traditional students (having served in the milit ary for several years), and have relatively generous tuition benefits through the Post 9/11 GI Bill that can make them attractive students for colleges and universities that are highly dependent on tuition. Finally, the Obama Administration has implemente d regulations for colleges and universities that will limit the federal financial aid dollars they can receive if their graduates fare poorly after earning a career focused degree (U.S. Department of Education, 2015). These regulations are aimed at reducin g the number of graduates from low performing for profit schools who pay substantial tuition but are not able to find jobs that pay sufficient salaries to repay their debts (U.S. Department of Education, 2015). T hese examples focus on concern about differ ential outcomes between public and non public colleges and universities. This concern is reflected in the popular media as well and tends to focus on the tuition prices for private non profit and private for profit

PAGE 21

10 schools, which are usually higher than st ate run public colleges and universities (College Board, 2013; Jacobs, 2013; National Center for Education Statistics, 2013; Schlesinger, 2010). Additionally, policymakers have raised concerns that those who attend private for profit colleges and universit ies have lower graduation rates and higher student loan default rates, indicating that they may be earning less than graduates from other sectors ( U.S. Senate Health, Education, Labor, and Pensions Committee, 2012 ) The counterargument is that these outcomes are primarily due to selection bias and these institutions tend to enroll low income students, or those who are less well prepared academically and would have lower earnings no matter the sector from which they graduated (Deming, Goldin, & Katz, 2 011). Although some research has examined the issue for first time students, little, if any, attention has been paid to how public, non profit and for profit colleges and universities may differ in how they serve returning students and whether the outcomes by near completers may differ. In addition to the statewide degree completion programs discussed above that try to attract near completers to public schools, many private non profit and private for profit colleges and universities are also engaged in similar programs (see for example Bellevue Unive rsity, 2013; University of Phoen ix, 2015 ). There has been no research to show the sectors in which near completers are most likely to enroll. D ata on adult enrollment however, can shed some light on the ques tion ( assuming that most of these students fall ay from postsecondary education) Those e nrollment data show that a large number of adults are enrolling in private schools (see Table 1 ) These data may not perfectly fit the population of interest, but they do suggest further study is warranted into the outcomes of near

PAGE 22

11 completers by sector. Based on these data, it appears that enrollment patterns may shift in relation to age, with older individuals shying away from private, non profit schools and being more likely to attend either public or private for profit ones. Although this is just an approximation of enrollment rates for near completers, r esearch that can identify whether the outcomes for those students vary by institutional type would contribute significantly to the ongoing policy discussions surrounding public and private schools. Table 1 : College e nrollment by age group and sector, 2011 Age Group Pu blic College or University Private, Non profit College or University Private For profit College or University Under 24 59.7% 28.2% 12.0% 25 29 69.0% 18.7% 12.4% 30 39 64.5% 16.6% 18.9% 40 and over 64.0% 16.2% 19.8% Source: National Student Clearinghouse, 2012a, 2012b Finally, although there has been some limited market research that examines the types of marketing and outreach messages that appear to resonate with near completers, there has been little formal research into the characterist ics that are associated with returning to finish a degree (see for example EducationDynamics, 2010; Maguire and Associates, 2010; Minnesota State Colleges and Universities, 2013). Better understanding the individual characteristics that may lead near compl eters to return to postsecondary education, as well as those that may prevent them from returning, could help policymakers better design programs to serve this population. Practical Implications Conclusion The research that follows canno t fill all of th ese gaps in the policy environment surrounding near completers. However, by examining the economic returns of near

PAGE 23

12 completers, the individual characteristics that may affect those returns the educational outcome s of near completers by sector and the factors that affect the likelihood of a near completer finishing a degree, this study contribute s greatly to ongoing efforts to increase the attainment rates of this group. Research Q uestions Thus, while there is a strong policy argument behind t he need to increase degree completion by near completers there is a need for rigorous research to substantiate many of the assumptions underlying such programs. State governments, as well as colleges and universities are committing millions of dollars to programs aimed at serving these students, while large numbers of college stop outs are spending tens of thousands of dollars and sacrificing substantial time to finish degrees all based on these untested assumptions. Based on the gaps discussed above, this dissertation will examine the following questions : RQ1: Do near completers who return to finish a baccalaureate degree earn a positive economic return compared to near completers who do not return ? RQ2 : Does the sector of the college or university from which a near completer graduates affect his/her economic return? RQ3 : How do the characteristics of individual near completers affect the likelihood that they will return to a college or university and complete a baccalaureate degree? Answering these research questions will make contributions to higher education and workforce development practitioners by evaluating a key claim of a major public sector effort and by providing more information about the types of students that current

PAGE 24

13 effo rts are successfully reaching As detailed more fully in the following section, this study will also contribute to academic research in labor economics and public administration. Hypotheses for each research question are derived from the review of li teratu re and are presented in Chapter 2

PAGE 25

14 CHAPTER II LITERATURE AND HYPOTHESES The literature that informs the se research questions comes from different fields. The basic question of whether a near completer who returns to finish a baccalaureate degree is likely to earn a positive economic return relies significantly on human capital and signaling theories from the field of labor economics. The second research question draw s on public management literature that focuses on the relationship between public and private management of an organization and outcomes of public programs, as well as literature on outcomes by sector. This leads to a review of literature on the outcomes of students at public, non profit, and for profit colleges and universities. 4 The final research question draw s on the limited research about the factors that are associated with individuals (particularly adults and non traditional students) pursuing additio nal education and training. The literature r eview is structured accordingly, beginning with an overview of research on returns to education before focusing more specifically on returns to adult education and the limited studies available on the returns to near completers. Following this, the review examines potential sources of bias in research on returns to education and methodologica l approaches to addressing them This discussion includes an examination of the factors that affect decisions to pursue add itional education and training. The review concludes with an examination of the research on the outcomes produced by public and private organizations generally, and more specifically, the existence of research on 4 Although traditionally in higher education literature these sectors are referred to as public, private, and for profit (or proprietary), to be consistent with public administration lite

PAGE 26

15 differential outcomes between public non p rofit, and for profit colleges and universities. As noted above, this study focuses exclusively on near completers who return to complete baccalaureate degrees. However, the literature review draws on studies on all types of postsecondary credentials due t o the limited extant research on the population of interest. The Economic Return to Degree Completion Labor market economics as a field has developed an immense body of literature related to the individual financial benefits of education This section of t he literature review begins with a discussion of the foundations of human capital theory and the empirical model that forms the basis for estimates of the impact of schooling on wages followed by a discuss ion of the theoretical origins of signaling theory and t he so called The review then turns to research specific to the economic returns to college completion, followed by a review of the limited research on degree completion by near completer s. This section finishes with a discussion of methodological refinements in the literature that have helped to reduce potential sources of bias in estimating the economic benefit of additional education. Human capital t heory. Literature on human capital theory forms the foundation for answering the question of whether near completer s who return to finish a degree will receive an economic benefit. Mincer (1958) and Becker (199 3 [1964]) are generally viewed as providing the key beginning points for human ca pital theory even though both authors build on a number of predecessors in economics literature (see for example Moore 1970 [1911]; Staehle, 1943). Mincer eponymous equation has become the standard method of estimating economic returns on schooling.

PAGE 27

16 earnings rise with age and experience, but then begin to decline at older ages ( Mincer, 1958). T he amount of schooling and training individual s receive also affects their beginning salaries and impact s the slope of lifetime earning s meaning that those with more schooling are likely to see their income continue to increase relative to those with lower levels of education (Mincer, 1958) The standard Mincer equation is as follows: ln Y = ln Y 0 + 1 s + 2 e + 3 e 2 Where Y is an 0 is the earnings of an individual with no schooling or experience, s represents years of schooling, and e represents work experience (Mincer, 1974). The quadratic experience term accounts for the changes in the earnings function as experience grows. Mincer also came to include a vector of individual characteristics that also affect earnings including racial /ethnic background, sex, family status, and characteristics of the city in which the individual lives (Mincer, 1958, 1974). Whi le this equation addresses the relationship between earnings and a variety of factors, it relies on a series of ass umptions in order to provide an actual rate of return rather than just the wage premium for an additional year of schooling The equation ass umes that there are no direct costs associated with schooling, that an individual will work the same length of time independent of his or her schooling level, that education and experience can be estimated separately (i.e. they do not have a combined effec t on earnings) and that there are no income taxes (Ashenfelter, 1978; Bjorklund & K j ellstrom, 2000; Card, 1999; Heckman, Lochner, & Todd, 2008 ; Heckman Lochner, & Todd, 2006). The Mincer equation also focuses on years of schooling and assumes a linear relationship between years of schooling and the log of earnings, without deviations from that linear path that can be evident for completing certain credentials like high

PAGE 28

17 school d iplomas and college degrees (i.e. the 12 th and 16 th years of schooling) (Card, 1999; Heckman, Lochner, & Todd, 2008). Rates of return estimated using a Mincer equation generally do not incorporate discount rates because they apply equally to earnings for b oth the treatment and control levels of education, so the discount can be dropped from consideration. Additionally, there are several difficult to measure variables omitted from the equation, such as innate ability, ambition health, and parental education that can bias estimates although the amount of bias may be quite small (Becker, 1993). Discussion of ability bias and selection bias in studying the returns on education continues to the present day and is presented in greater detail below equation, with its assumptions, is a special case of a more general equation incorporates costs, and because costs are borne at the time education is received, he also incorpora tes a discount ra adapted from Heckman, Lochner, and Todd (2006) is as follows : In this equation (with the explanation adapted for the purposes of this study) Y 1 represents the earnings for an individual with a near completer who has finished a baccalaureate degree at time t while Y 0 represents the earnings for a near completer at time t; r represents a discount rate, while represents the number of years an individual is enrolled in college The first numerator term sums the total earnings starting at t= which means it only counts earnings after schooling is complete. T he second numerator term (and the identical denominator) sum the total earnings from t=0. The practical effect of

PAGE 29

18 this is that an enrolled individual earns nothing while pursuing a degree The direct costs for attending college are included in the term C t where C represents direct costs per year to complete a degree. Once the individual has finished his or her degree this term equals zero. importance of including direct costs. Under this formulation, the economic return for a near completer finishing a degree is positive if his or her wages (appropriately discounted), subtracting costs of attendance are greater than the wages of a near completer who does not return to school. Although t his equation does not account for the possibility that near completers who are working on degrees may also be working at the same time, it forms a more complete start ing point for determining whether there is a positive return for degree completion. The models proposed in Chapter 4 combine aspects of both approaches to address the key research question about returns to degree completion. Wage premiums and postsecondar y education. There is strong agreement within the literature that additional years of schooling are associated with increased earnings using a variety of methodologies, data sources, and approaches (Ashenfelter & Rouse, 1998; Ashenfelter, Harmon, and Oost erbeek, 1999; Hausman & Taylor, 1981; Hout, literature review will not focus on years of schooling in detail, but rather the wage premium for degree completion.

PAGE 30

19 The emp irical evidence that individuals receive a wage premium for completing a college degree is also relatively strong (Cellini & Chaudhary, 2014 ; Grubb, 1997; Jepsen, Troske, & Coomes, 2014; Kane & Rouse, 1995; Marcotte, Bailey, Borkoski, & Kienzl, 2005). The estimates of the premium vary somewhat across studies as researchers use different comparison groups, disaggregate data differently, and examine different subgroups of interest. An overview of the some of the estimated returns is included in Table 2. Table 2 : Returns to degree completion Degree Level Grubb (1997) Kane & Rouse (1995) Marcotte, et al. (2005) Male Female Male Female Male Female Assoc Degree 18.1% 22.8% 26.6% 20.7% 17.1% 40.4% Bach Degree 54.8% 53.4% 52.5% 39.2% 45.8% 91.9% Above Bach. See Below 95.2% 53.1% N/A N/A 64.9% 77.9% N/A N/A N/A N/A Prof. Degree 174.6% 153.7% N/A N/A N/A N/A Ph.D 122.6% 141.3% N/A N/A N/A N/A professional degree, and Ph.D, separately. Kane and Rouse group these together. Marcotte, et al., include only associate b achelor s degrees. Wage premiums are not the only economic benefit that those who finish college degr ees receive. Degr ee holders also have lower unemployment rates and greater buffering through economic downturns (Gangl, 2006; Hout, 2012; Hout, Levanon, & Cumberworth, 2011; Jepsen, Troske, & Coomes, 2014). These additional effects could form a meaningful portion of the ov erall monetary benefit that an individual receives for returning to complete a postsecondary degree, depending on how wages are calculated and what controls are used for employment. One estimate suggests that 2/3 of the return to education is due to higher wages, while 1/3 is due to more hours worked (Card, 1999). This suggests that using annual wages as an outcome variable may be prudent to capture the full effect of education on earnings.

PAGE 31

20 Heterogeneous effects of degree completion. The benefits of degree completion may not accrue to members of different sub groups equally. The effects of education may be heterogeneous across certain important characteristics such as racial/ethnic background, socio economic status and gender. Individuals of lower socio ec onomic status may actually benefit more from postsecondary education than those from more advantaged backgrounds, as research shows that those who are least likely to graduate from college realize a 30 percent wage gain, while those who are most likely to graduate realize a 10 percent wage gain (Brand and Xie, 2010). These findings counter the idea of positive selection bias, which suggests that those individuals who are most likely to earn higher wages anyway are also most likely to attend college (see for example, Carneiro, Hansen, and Heckman, 2003). Othe r research finds different outcomes by gender, racial/ethnic background, number of children, and other factors (Bailey, Kienzl, & Marcotte, 2004; Bitzan, 2009; Henderson, Polacheck, & Wang, 2011; Jaeger & Page, 1996). This suggests that the individual characteristics of near completers may affect their economic return. Signaling theory. Signaling theory holds that education level serves as a signal to potential employers about his or her a bilities and characteristics, such as potential productivity, work ethic, persistence and more ( Arrow, 1973; Weiss 1995 ). Individuals pursue a level of schooling commensurate with these abilities, thus indicating their relative value in the labor market. Higher incomes, then, may not be due to the additional years of schooling, but to the Under signaling theory, e mployers use education as a screen to hire the most desirable workers those who are less likely to be unhealthy (and thus miss work), less likely to

PAGE 32

21 leave the job, and more likely to avoid habits such as excessive drinking and drug use provided these characteristics are correlated with the amount of schooling an individual possesses (Weiss, 1995) Al though related to human capital theory, this presents a different argument in that individuals obtain a certain amount of education to send certain signals to employers, rather than improving their potential productivity through learning. Signaling theory effect. This is a phenomenon whe re a particular educational credential (conferred through would be expected from an equivalent amount of schooling. As an example, someone who completes 16 years of school but no college degree would be predicted to earn less than someone with the same amount of schooling but has received a degree. Research has confirmed that the sheepskin effect exists (empirical results are discussed in greater detail below) supporting signaling theory and accounting for a portion of the wage premium received by college graduates although this effect does not appear to apply to all groups equally A related view is that individuals considering how much education to obtain may make this decision based on their own views of their earning potential so that they can accurately signal this potential without incu rring unnecessary extra costs in t uition or foregone earnings (Carneir o, Hansen, and Heckman, 2003). Based on this view, workers who feel that their jobs are not appropriately matched to their abilities and characteristics could try to finish a degree in o rder to send a different signal to their potential employers. Empirical evidence for the sheepskin e ffect. While much of the research cited previously about the wage premium for college completion does not specifically test for

PAGE 33

22 the sheepskin effect, stud ies that do explicitly examine this phenomenon provide further evidence that individuals who earn college degrees receive a wage premium. Controlling for years of schooling, several studies show that college diplomas generate a wage premium for individuals ( Bitzan, 2009; Hungerford & Solon, 1987; Ja e ger & Page, 1996; Jepsen, Troske, & Croome, 2014; Park, 1999) Other results for estimations of the Table 3 : Estimations of the sheepskin effect Author(s), year Findings Bailey, Kienzl & Marcotte, 2004 Insignificant sheepskin effects for males at all degree levels; 28% wage insignificant for other degrees. Belman & Heywood, 1991 21% wage premium for degree completi on by racial/ethnic minority males, compared to 10% premium for white males. 25% wage premium for degree completion by racial/ethnic minority females, compared to statistically insignificant premium for white females. For lower level signals (i.e. grade sc hool and high school) racial/ethnic minorities had lower estimated sheepskin effects than white individuals. Bitzan, 2009 Black males receive higher premiums for graduate degrees than white males. Hungerford & Solon, 1987 Jaeger & Page, 1996 Earlier estimates of sheepskin effect are biased because data did not include both degrees and years of schooling, but relied on imputing degrees. statistically significant differen ces between effect s on different sub groups. insignificant for black males. 27 statistically insignificant for bla ck females. Jepsen, Troske, & C r oome, 2014 the diploma. Kane & Rouse, 1995 Small evidence of other combinations. Park, 1999 compared to those with 14 years of schooling but no degree.

PAGE 34

23 sheepskin effect are mixed, with only some subgroups receiving a benefit or findings of no effect (Bailey, Kienzl, & Marc otte, 2004; Kane & Rouse, 1995) A comparison of results is presented in Table 3 T here is general agreement that the sheepskin effect is real and does provide a wage premium to some of those who complete degrees but it may not be present for all subgroups Some of the differences in the estimated size of the sheepskin effect are likely due to differences in selected data and methodological approaches Research only focused on identifying non linearities in the overall return to schooling for parti cular years in which individuals earn diplomas or credentials (for example, the 12 th and 16 th years of schooling) may be overestimating the sheepskin effect (Skalli, 2007). Estimates for the size of the sheepskin effect that incorporate both years of schoo ling and actual degrees earned are lower than those that only use years of schooling (Bitzan, 2009; Jaeger & Page, 1996). Still, even with the differences in estimates, the existence of the sheepskin effect could suggest that near completers who finish a d egree are likely to receive a wage boost even if they only need a small number of credits to finish a degree. Heterogeneous sheepskin e ffects. As can be seen in Table 3 some r esearch on the sheepskin effect also finds h eterogeneous returns for different population sub groups. Belman and Heywood (1991) find that women and racial/ethnic minorities have greater sheepskin effects than white males. Ja e ger and Page (1996) however, suggest that this finding could be biased and that with proper measurement of degree completion, there is no evidence that there are differential returns to diplomas. Kane and Rouse (199 5) find only a small sheepskin effect for men completing a baccalaureate degree and women ult for other combinations. They

PAGE 35

24 due to the earning potential of nursing degrees (Kane & Rouse, 1995) Bitzan (2009) finds that white males receive larger sheepskin effects than black males e finds that black males receive a larger sheepskin effect (Bitzan, 2009). While this research is not focused specifically on near complet er s, it does suggest that estimates of wage premiums due to diplomas may differ by gender and racial/ethnic background. Wage premiums for degree c ompletion by near completers There is limited literature that examines the economic benefit of near complete rs who finish degrees. There are some case studies examining the outcomes of students who finish degrees through a coll ege or university that offers a degree completion program for stopouts The population under consideration bears some similarities to nea r completers studied here, but few studies specify the population beyond examining college stopouts. These studies generally rely on post completion surveys of program completers. Mishler (1983) surveys graduates after stopping out and finds large numbers reported job related improvements due to earning their degree s Green, Ballard, and Kern (2007) use a similar approach to survey graduates who returned to complete d egrees and again find evidence for positive career outcomes based on degree completion. These findings are echoed by a series of methodologically similar evaluations (Culver, 1993; Harris, 2003; Hoyt & Allred, 2008; McKinney, 1991). These evaluative effort s are suggestive but do not employ relevant comparison groups and thus differ significantly from the study carried out here

PAGE 36

25 Wage premiums for adult education and training. Another relevant thread in liter ature on wage premiums for education compares what adults who pursue education later in life might earn compared to those who complete their education and training earlier, such as those students who attend and complete college directly after high school. s for adults who earn credentials are smaller (and in some cases non existent) compared to those who ea rn degrees one a more traditional pathway 5 While their work includes those adults who already have some college credit, their inquiry focuses on the tim ing of educational attainment and does not focus specifically on near completer s. Additionally, they limit their estimation to hourly wages as an income measure, which could mask some benefit as research suggests one pathway through which education affects earnings is by increasing the hours an individual works (Blanden, Buscha, Sturgis, & Urwin, 2012; Card, 1999). Even with these caveats, t his research shows that comparing near completers who return to traditional students may not be appropriate and that t here are differences in returns to postsecondary credentials depending on when they are completed. Light (1995) examines the returns for those who stop out of education and return later compared to those who complete a similar amount of schooling but on a direct pathway. Those individuals who return to school after a period in the workforce experience w age gains, but they are less than those who complete their schooling on a direct path (Light, 1995). T his research does not focus on the level of education obtained when the indi vidual stopped out Leigh and Gill (1997) follow a similar path in 5 traditio

PAGE 37

26 comparing a dult community college graduates with traditional aged graduates and find adults actually recei ve a slightly greater benefit. There is also research from abroad that examines outcomes for adults pursuing any type of education. Research from Sweden finds th at individuals who complete degrees as adults have 18 percent higher employment rates and 12 percent higher earnings (Hallsten, 2012). Other European evidence also suggests that adults who pursue education may end up with better employment prospects (Kilpi Jakonen, de Vilhena, Kosyakova, Stenberg, & Blossfeld, 2012). These comparisons examine adults who pursue additional education in comparison with adults who do not, which is a more relevant approach for the current study (Hallsten, 2012). T here has been significant research on wage benefits received by adults who finish additional education, but little of it focuses on near completers and none of it focuses on whether their decision to return to finish a degree is likely to result in a positive economic r eturn Although adults are not directly comparable to near completers, t he findi ngs above do suggest that near completers who finish degrees may receive an economic benefit Without estimations of foregone wages and direct costs and without comparisons to other near completers who do not finish degrees it is not possible to conclude from the literature that they are likely to receive a positive economic return Heterogeneous effects for adult learning. As noted above, researchers have identified heterogen eous effects of educa tion on different sub groups. T his is an important factor in examining wage premiums for adults as well. Brand and Xie (2010), as noted above, find strong evidence that individuals with lower socio economic status benefit more from pos tsecondary education than those from advantaged backgrounds.

PAGE 38

27 degrees. They find a statistically significant wage premium for women who complete for men. Other researchers have also found evidence of differential returns by gender for adults participating in education and training programs (Blanden, et al., 2012; Hallsten, 2012). Blanden, et al. (2012) find that women who complete credentials as a dults receive a 10 percent wage premium while similar men receive no premium. Similar to sheepskin effects, this may be due to different credentialing requirements in fields in which men and women tend to work (Blanden, et al., 2012). These findings, coupl ed with research on heterogeneous sheepskin effects by gender and racial/ethnic background, suggest that economic returns for near completers who finish degrees may differ by sub group membership. S o urces of bias in estimates of the return to e ducation Human capital and signaling literature also recognize key methodological considerations in estimating the crucial factor in determining whether increases in earnings are characteristics or the schooling or training received (Becker, 1993). Highly able individuals may be more likely to pursue additional schooling as well as earn high wages regardless of schooling so estimates that do not control for ability will be upwardly biased (Becker, 1993; Card, 1999; Griliches, 1977). Many researchers believe, with some empirical support, that although ability bias exists and leads to overestimations of the return to education, the amount of bias is relativ ely small (Becker, 1993; Card, 1999; Griliches, 1977). This section examines selection bias in general research on the returns

PAGE 39

28 to education before turning specifically to issues surrounding this potential bias in adult education and training programs. Effo rts to mitigate selection b ias. Extensive theoretical and empirical work has sought to eliminate ability bias as a concern in estimates of the return to education The original approach was to include control variables for measures of individual ability i n an ordinary least squares (OLS) estimation. Including standardized test scores from high school as a control variable has been o ne approach, although this may not be fully effective (Card, 1999). Additional controls for family background, high school ach ievement, and demographic characteristics are also regularly used. However, c oncern that these variables may not fully capture innate ability and motivation, thus not eliminating ability bias, has led researchers to employ a range of other methodo logical a pproaches to develop more accurate estimate s for the effect of schooling on earnings. Although it is not feasible to use experimental controls to assign individuals randomly to higher or lower education groups to test for ability bias, researchers have tri ed to approximate such conditions. One such approach has been to study the wages of identical twins with different levels of education to attempt to quantify the size of ability bias in OLS based estimations. This rests on the assumption that identical tw ins have the same innate ability and family background, so if they have different levels of education, any differences in their earnings would be attributable to schooling. Behrman and Rosenzweig (1999), following this approach, estimate that ability bias leads to an overestimation of the impact of additional schooling on earnings by 12 percent. Card (1999) carries out a meta analysis of samples using twins and finds a roughly similar bias of about 10 percent.

PAGE 40

29 However, two other methodologies that aim to c ontrol for ability bias instrumental variables and fixed effects models have consistently resulted in larger estimates of the economic benefit of education compared to OLS estimates that attempt to control for bias through observed variables 6 Hausman and Taylor (1981) find a seven percent wage premium for each additional year of schooling using OLS compared to a nine percent wage premium using an instrumental variable. Ashenfelter, Harmon, and Oosterbeek (1999) compare different methodologies and find that instrumental variables and fixed effects lead to near identical results. Their meta analysis suggests that the wage premium of an additional year of schooling using instrumental variables or individual level fixed effects is about nine percent per yea r of schooling, compared to seven percent with OLS based approaches using control variables to account for bias Other research supports these findings, concluding that using individual level fixed effects models produces estimates that are between 40 and 50 percent larger than other approaches (Ashenfelter & Rouse, 1998; Ashenfelter & Krueger, 1994). These findings have generated divergent explanations. Some suggest that this shows evidence of negative selection bias, which holds that those who are least likely to complete additional schooling actually benefit the most. Brand and Xie (2010) conclude research must examine heterogeneous returns to education namely that different sub groups (particularly low income individuals and racial/ethnic minorities that may be less likely to complete higher levels of education ) receive different economic benefits from 6 Both of these approaches attempt to mitigate bias due to omitted variables and approximate experimental conditions. Fixed effects approac hes, which are used in this study, are explained in greater detail in the chapter on methodology. Instrumental variables try to correct for situations where independent variables are correlated with the error term by identifying an instrument that is corre lated with the independent variable of interest but not the error term. For additional information, see Wooldridge (2006) and Angrist and Pischke (2009). For the purposes here, the key fact is that these approaches can, in theory, greatly reduce concerns t hat innate ability or other omitted variables are leading to bias in estimates of the economic return to education.

PAGE 41

30 a dditional learning and credenti als. Others argue that OLS approaches using control variables without instrumental variables or fixed effects may underestimate the wage premiums for education due to measurement error that can negatively bias coefficients (D eaton, 2010; Hout, 2012). An alternative explanation focuses on instrumental variables and suggests that there could be publication and reporting bias as researchers are most likely to select instruments that produce statistically and substantively signifi cant results (Ashenfelter, Harmon, and Oosterbeek 1999). With appropriate data sources, howeve r, individual level fixed effects approaches control for all time invariant characteristics of an individual (Wool d ridge, 2006). Thus, estimations based on fixed effects models must true, they will properly account for ability bias. The chapter on methodology discusses these characteristics in greater detail. Selection bias a nd near completers One important question in e xamining whether near completers who finish degrees receive a wage premium is whether those who finish are substantively different from those near completers who do not return. It is possible that these diffe rences, whether in work ethic, demographics, ability, or other characteristics, may affect income in ways that could mistakenly be attributed to the effect of degree completion. Again, with limited extant research on near completers examining the availabl e research on adult learners can be a helpful starting point. Motivation for pursuing additional education is one consideration. If adults are pursuing additional education for non economic means, it could be a factor if no wage premiums are evident. Alte rnatively, it could be that those who are certain they would receive workplace benefits if they complete a degree are much more likely to return and

PAGE 42

31 finish their schooling. Houle (1961) conducted the foundational study on the motivation of adult students and identifie d three major types of adults pursuing education: goal oriented individuals, activity oriented individuals, and learning oriented individuals. Other research has identified professional advancement as well as pursuing learning for its own sak e and enjoying participation in group activities as reasons that adults pursue additional education (Morstain & Smart, 1974). Further work has confirmed and augmented H and concludes that adults pursuing education consist of those who are economically motivated as well as those who are motivated by the enjoyment of learning or desire to participate in group activities (Boshier & Collins, education and lin king that to income data with available datasets However, this could be an important confounding factor in an analysis of near completers. There could be a relationship between the motivation for pursuing education and resulting income that is unrelated t o the effect of additional schooling on income. For example, an individual who becomes more ambi tious later in life may dec ide to finish his or her degree and later earns higher wages due to that increased ambition rather than the degree he or she finished It may be difficult, however, to separate the effect of education from changes in ambition. Other research has sought to identify some of the observable factors associated ily background and socio additional education. Additionally, they find that having children at an early age, entering the military, and belonging to certain racial/ethnic groups are also positiv ely associated

PAGE 43

32 explanation of these latter factors could be that individuals who had early children or entered the military were more likely to leave education short of their desired attainment. Age, marital status, family size, family income, and tuition levels are also important factors in predicting adult enrollment (Light, 1996). While these factors appear to be important for adult enrollment, further research is necessary to determine whether they are applicable to near completers, as well. The Ashenfelter d ip. In considering the current study, one source of potential bias is the income level of near completers prior to reenrollment. A large body of research suggests that for both education and training programs, lower pre treatment incomes are a significant factor in predicting participation. These findings come from studies examining both postsecondary education and workforce training programs (Ashenfelter, 1978; Cellini & Chaudhary, 2014 ; Heckman & Hotz, 1989; Jepsen, Troske, & Coomes, 2014 ) Ashenfelter (1978) originally identified the phenomenon that now bears his name. The program s tends to decline immediately prior to the treatment. Following completion of the training program, as tends to naturally recover, est imations of its impact may be overstated as this natural recovery is interpreted as a treatment effect (Ashenfelter, 1978). This phenomenon may also be evident across a population during periods of economic recession, in which adults may be more likely to seek out training and education programs I mproved individual incomes following recessions are likely due in part to natural economic recovery in addition to any benefits from the training and education programs (Ashenfelter, 1978).

PAGE 44

33 Others have found evidence of the Ashenfelter dip in training a nd education programs. Heckman and Smith (1999) find a pre training decrease in income for some that participate in training programs. Marcus (1986) finds that adults whose earnings decline below values predicted from relatively straightforward regression models are more likely to return to school. Blanden, et al., (2012) using data from England conclude that credentials earned later in life (including the equivalent of U.S. high school degrees, vocational certificates, and college degrees) have economic benefits for women but find that the effect for men disappears when measures are taken to control for trends in pre treatment income Their work is slightly different in that they identify a dip followed by the beginnings of an upward income trend for adu lts participating in education and training programs for men, but not for women. Cellini and Chaudhary (2014 ) and Jespsen, Troske, and Coomes (2014) also find evidence of an Ashenfelter d ip for adults who pursue community college degrees. Additional resear ch has concluded that pre treatment income is an important determining factor in whether adults enroll in further education, with lower income adults more likely to enroll (Jepsen & Montgomery, 2012; Light, 1996). Thus, it will be imperative to account for the income of near completers prior to their return to postsecondary education. If there is an Ashenfelter Dip, it would potentially bias estimates of the wage benefit for degree completion. Foregone wages and indirect costs. As described above, accoun ting for foregone wages is a central part of determining whether an individual receives a positive economic return to degree completion. However, f oregone wages could also be a source of potential bias in the estimates of the wage premium for degree comple tion. If returning students forego substantial wages while enrolled, their incomes would recover naturally when they

PAGE 45

34 start working more after completing a degree which could upwardly bias estimations. Again, with no literature available specific to near c ompleters, I turn to research on adult learners instead. The literature on whether older students forego significant wages is somewhat mixed. Jespsen and Montgomery (2012), who found that lower income individuals are more likely to seek out additional educ ation and training argue that this shows that opportunity costs are a significant barrier for adults to enter postsecondary education They suggest that those with higher incomes would pay higher opportunity costs to complete additional education (Jepsen & Montgomery, 2012) However, they present little data to conclude that when adults reenroll in postsecondary education they forego earnings. An alternative explanation could be that those adults who are already faring relatively well economically see less need to pursue additional education. Blanden, et al. (2012) suggest that evidence of depressed wages prior to completing adult education and training programs could be due to foregone wages or simply that productivity is lower for these adults. E mployment status prior to education has also proved to be an important factor in assessing benefits of adult education in multiple European countries, and studies there have shown some evidence that adults may forego wages to pursue additional education (K ilpi Jakonen, et al., 2012). Thus, while initial salary may be a key determinant of decisions by adults to enter postsecondary education, the literature suggests that adults forego wages to pursue additional education and training. However, there are few studies that examine adult college going in detail, and it could be that programs are more flexible and allow these students to keep working full time while enrolled. Additionally, data show that adults tend to opt for part time enrollment and other flexib le options offered by higher

PAGE 46

35 education institutions ( National Center for Education Statistics, 2014 ). Still, it will be important to control for foregone wages so that any natural recovery following school does not appear to be a benefit of degree completi on Additionally, estimations of these foregone wages are necessary to account for the full cost of returning to finish a degree. Distinctions between Public and Private Organizations A central question in public administration and public management resea rch has 1987; Walmsley & Zald, 1973). The roots of the focus in public administration on the differences between public and private management are evident in the work of m any of the foundational writers in the field. This section of the literature review approach es the question of public and private organizations by first examining the criteria used by the field to delineate the two types of organizations, then illustrating how an operational definition of public and private management has been used in research on higher education, before concluding with a discussion of the literature on inputs and outcomes at public and private colleges and universities. Foundations of pub lic vs. private debates. Perhaps starting with ( 1967 [1912 ] ) principles of scientific management, which were originally developed for private firms the tension between public and private organizations is evident in public administration That came to be applied to distinctly public government agencies suggests that early in the field, there was a blurred distinction between the two sectors Weber ( 1970 [1922] ) in his classic treatise on bureaucracy, argues that his conclusions ap ply equally to the government and private sectors. Simon (1957) suggests blurred lines between public and private organizations, arguing that assumed distinctions

PAGE 47

36 may not apply within actual organizations. Dahl and Lindblom ( 2000 [ 1953 ] ) proposed a continu um of public and private organizations, arguing that it is not feasible to identify distinctly private or public organizations. While these foundational theorists suggest difficulty in drawing specific distinctions between public and private organizations more recent research has attempted to clarify the differences. T he focus on which type of organization can produce better outcomes (and in what circumstances) grew in response to a major effort by governments in the United States, Europe, and elsewhere t o decentralize government authority and privatize government services under the heading of New Public Management (Hood, 1991). This emphasis on determining whether public or private organizations produce better results is a key field of research in public administration, however, it starts with an assumption that there are clear, consistent, and easily identifiable criteria to distinguish the two types of organizations. What makes an organization public or p rivate ? A significant vein of the literature has focused on the criteria that differentiate organizations in the two sectors D oing so can be difficult and researchers have used different approaches, including placing organizations on a public private continuum rat her than drawing strict distinctions (Bozeman, 1987; Moulton, 2009; Rainey, Backoff, and Levine, 1976; Rainey & Bozeman, 2000). Criteria can include: involvement (or not) in the economic market for revenues and resources; the extent of formal legal constra ints placed upon the organization; the intensity of outside political influence on the organization; the extent of the organization; the extent of public scrutiny and public expectations; the type of goods

PAGE 48

37 produced (public, quasi public, or private); the criteria for evaluation; the extent of fragmented authority; the emphasis placed on performance; and the incentives offered to workers (Rainey, Backoff, & Levine 1976 ). A more succinct set of criteria focuses on who owns the organization, how it is funded, and what type of social control is exerted upon it (Bozeman, 1987; Hvidman & Andersen, 2013; Perry & Raine y, 1988; Walmsley & Zald, 1973). However, even these relati vely straightforward criteria can lead to To further simplify the concept, some researchers have used the relatively straightforward criterion of regulatory authority and owne rship to determine whether an organization is public or private (Bozeman & Bretschneider, 1994 ; Feeney & Welch, 2012; Gibson, 2011; Monks, 2000 ). Thus, throughout public management literature, there are a variety of operational definitions for determining what makes an organization public or private. The field has not reached consensus on the best way to approach the topic However, research focused on certain sub fields, such as postsecondary education, has identified usable distinction s b etween public and private organizations. Public and private in higher e ducation. university would be somewhat difficult based on the numerous criteria delineated above Based on the definition of whether or not an organization pursues public purposes, colleges and universities could all be considered public ( Using main sources of revenue could also be problematic. A lthough nominally public colleges and universities all receive some government funding, some research suggests that those

PAGE 49

38 that have high tuitions may behave more like private colleges and universities particularly when they operate in a decentralized state gover nance regime (Knott and Payne, 2004) Further, as direct g overnmental funding of publicly owned colleges and universities has declined, research suggests they have become more like privately owned schools (Duderstadt and Womack, 2003). Additionally, given that many privately owned colleg es and universities derive significant revenue from tuition paid by federal grants and loans as well as receive significant public revenue in the form of federal research grants they certai nly could be considered public entities under some of the criteria above ) Within higher education policy and research circles, however, colleges and universities are regularly classified as public or private based on political control and funding es sent ially the concept of ownership used elsewhere in public management and higher education literature and research ( see for example College Board, 2013; Feeney & Monks, 2000; National Center for Education St atistics 2013 ; Tierney, 1980 ) S chools that are overseen by public agencies or entities and funded directly by state governments are classified as public, while those that are not are classified as private. Within the label of private, there are for profi t and non profit schools, w hich are categorize d based on their legal status (Cellini & Chaudhary, 2014 ; Deming, Goldin, & Katz, 2011) These classification s are incorporated in the Carne gie classifications, which are used throughout the higher education policy and research communities. 7 While these definitions ignore some of the nuance identified elsewhere in studies on publicness, they do provide fairly straightforward categorization that is amenable to empirical research. Further, adh ering to consistent definitions across 7 For additional information on the Carnegie Classifications, please see http://carnegieclassifications.iu.edu/

PAGE 50

39 research efforts is crucial for developing meaningful and comparable results (Me ier, Publicness as an independent v ariable. R esearch looking at the outcomes of public, private non profit, a nd private for profit schools must address issues of selection bias. Studies show that the student populations vary significantly by sector Students at private for profit schools are more likely to be racial/ethnic minorities, tend to have lower incomes u pon entering, and are more likely to be adults than students attending schools from one of the other sectors ( Cellini & Chaudhary, 2014 ; Deming, Goldin, & Katz, 2011). Other research disagrees, showing that students at public colleges tend to have lower in comes and are more likely to be adults and part time students than those at private schools ; the contradiction may be due to the fact that this research does not capture recent growth in the for profit sector (Scott, Bailey, & Kienzl, 2006) Research using this operational definition of publicness includes studies that examine a variety of dependent variables, including student demand for higher education output by faculty, and overall economic returns to students ( Feeney & Welch, 2012; Monks, 2000; T ierney, 1980 ). There are also s tudies on graduation rates that show a range of findings, including that public schools outperform private non profit ones, that there is no significant difference, and that private non profit schools outperform public ones ( Gibson, 2011; ; Scott, Bailey, and Kienzl, 2006 ). None of this research examines privately owned for profit schools. Although federally reported graduation rates for private schools are, on average, higher, the strategy for c ontrolling for selection bias impacts findings (Scott, Bailey, and Kienzl, 2006). One important variable seems to be the amount of funding available per student. When it is

PAGE 51

40 included as a control variable, public schools outperform private ones (Scott, Bail ey, and Kienzl, 2006). Student characteristics and controls for ability bias also impact findings (Gibson, 2011; ; Scott Bailey, and Kienzl, 2006). Most of this research, due to its reliance on federal data collected on first time, full time students, explicitly excludes near completers. Other r esearch on the connection between student outcomes and the type of college or university at t ended has focused in recent years on the differences between outcomes for graduates of priva te for profit schools compared to graduates from private non profit and public schools. Examining outcomes, graduates from private for profit schools have higher loan default rates, lower earnings, and higher unemployment rates compared to graduates from o ther schools ( Cellini & Chaudhary, 2014 ; Deming, Goldin, & Katz, 2011). Other research has shown s tudents completing two year degrees at for profit schools have slightly larger economic returns than those from non profit schools, but this may be due to selection bias because many students who attend public and non profit two year schools subsequently pursue four year degrees which may negatively bias their immediate earnings after graduation (Lang & Weinstein, 2013). Additional research combines measures of school quality with sector type Research examining outcomes based on selectivity of the college or university and its public or private non profit status finds large economic returns for graduating from a highly selective private non profit school com pared to a low rated public school (Brewer, Eide, & Ehrenberg, 1999). Comparing public and private non profit schools of the same category shows no statistically significant earnings differential (Zhang, 2005). However research using more effective methodo logies to control for selection bias, such as

PAGE 52

41 matching similar students who attended different types of colleges and universities, as well as different quality levels, finds little difference in wages earned by graduates (Dale and Krueger, 2002). Other res earch shows slight earnings gain s for white students who finish degrees at private non profit colleges and universities compared to public schools, and no statistically significant earnings differential for non white students (Monks, 2000). In research exa mining gen der differences, Joy (2003) finds that women who attended private non profit schools that granted doctorates (a rough measure of quality) earned more than those who attended other private non profits and those that attended public colleges of eit her type (Joy, 2003) For males, there was no statistically significant difference (Joy, 2003). While these results show disagreements within the field of higher education research, they do not provide much guidance for the study at hand because they generally focus on traditional students and exclude near completers However, this research is informative because it shows that accounting for the characteristics of the students at schools in different sectors is crucial for understanding any differences in outcomes due to public, non profit, or for profit control. Thus, while there is consensus in the literature on the criteria used to differentiate public, non profit, and for profit colleges and universities, there is not agreement about earnings differ entials by school management type. Following the overwhelming majority of research into differences between outcomes at different types of colleges and universities, I use control of the institution, as defined by the widely accepted Carnegie classificatio ns. Although t here is limited research that focuses on near completer s who return to finish a degree and how the choice of school may affect their outcomes, research

PAGE 53

42 does generally show that there may be differential outcomes due both to student and school characteristics. Literature Review: Conclusion This review of literature spans many different topical areas. While there are few directly relevant studies focusing on near completers, research from labor economics and public management has identified nume rous critical issues for conducting a study of the economic return for near completers who finish degrees. Given the many different topics covered in the literature review, and the lack of a deep body of research on near completers, it is helpful to highli ght the important conclusions. Methodologically, the immense body of literature on the returns to education is nearly unanimous on the importance of accounting for potential selection bias. Bias due to ability as well income before a nd during enrollment co uld affect results if not properly controlled. Further, although studies may reach opposing conclusions on the benefits to certain sub groups, there is ample evidence that returns to education are not uniformly distributed by racial/ethnic groups and gende r. Additionally, research on outcomes when services are provided by public, non profit, and for profit sectors are not conclusive, esp ecially in the education realm. G iven the variations evident when looking at some outcome measures, further study is warra nted with regards to near completers. The research cited above informs the hypotheses I generate in the subsequent section, as well as the methodological approaches selected in C hapter 4

PAGE 54

43 Hypotheses This dissertation will contribute to policy discussions, to the academic literature on human capital and signaling theories, and to literature on the differences between public and private organizations by evaluating the four hypotheses. These hypotheses a re collected in Table 4 and linked to the original research questions that guide this dissertation Hypothesis 1 .1 : Near completers who return to postsecondary education to finish a baccalaureate degree will earn a positive individual wage premium compared to those who do not return The premium is sufficiently large to result in a positive economic return. The literature cited above supports the conclusion that on average, earning a wage premiums Those premiums tend t o be substantively large enough that most research does not bother accounting for direct and indirect costs of attaining the degree Given that near completers are much more likely to be older than students who proceed directly through postsecondary educat ion, they will have less time in the labor market to recoup costs Also, near completers who likely earn higher wages than students proceeding directly from high school through college may face higher indirect costs due to foregone wages. Additionally, with data showing that a substantial percentage of adult s enroll in privately owned for profit colleges and universities that cost more than public schools, it could be that, on average, returning near completers are paying higher direct costs than other types of postsecondary students, resulting in lower economic returns. Even with conservative estimates for direct costs, it is likely that at least some subgroups earn a positive economic return for baccalaureate degree completion.

PAGE 55

44 Hypothesis 1.2 : The benefits of degree completion by near completers do not accrue equally to all population sub groups The literature cited above suggests that there may be heterogeneous returns to education attainment based on subgroups disaggregated by gender, racial/ethn ic background, and socioeconomic status. It is likely that these differences will also appear in an analysis of near comple ters. Hypothesis 2 : Individuals who finish a degree at a public college or university will realize a higher economic return than tho se who finish a degree at a non profit or for profit college or university. While the literature is mixed on the differential returns to traditional students based on sector, data do show that non profit and for profit colleges and universities cost more t han in state tuition at public schools. In order to have the same overall return as those who attend lower priced public colleges, n ear completers who finish baccalaureate degrees at non profit and for profit schools likely must earn a higher wage premium for degree completion to offset the higher direct costs. However, given that there are not data on actual tuition prices paid by any individual, and that access to various forms of financial aid can vary significantly, it is important to understand that th ese approximated returns are based heavily on a series of assumptions about direct costs. An additional factor that could challenge this hypothesis is that for profit schools tend to emphasize the flexibility of their programs for adults, which could mean that graduates incur lower indirect costs in the form of foregone wages. Unless the results show higher wage premiums (or lower indirect costs ) for graduates of non profit and for profit schools, it is likely that their overall return will be lower. Grante d, such conclusions would be somewhat tentative due to the underlying assumptions that sources

PAGE 56

45 of financial aid do not vary by sector and actual tuition price paid by students is comparable to the average prices cited below. Hypot hesis 3 : Factors that are important in predicting educational successes for traditional students will also be important in predicting wh ether near completers finish degrees. Research on educational attainme nt suggests a number of factors that impact the level of education an indiv idual completes. These include socio economic status, racial/ethnic background, family characteristics, as well as unobservable characteristics such as motivation and ability. It is likely that many of the factors that impact traditional nal attainment will also apply to near completers. Table 4 : Research questions and hypotheses Research Question Hypotheses RQ1: Do near completers who return to finish a degree earn a positive economic return compared to near completers who do not return? H1.1: Near completers who return to postsecondary education to finish a baccalaureate degree will earn a wage premium compared to those who do not return The premium is sufficiently large to result in a positive economic retu rn. H1.2: The benefits of degree completion by near completers do not accrue equally to all population sub groups. RQ2: Does the type of management at the college or university from which a near completer graduates affect his/her economic return? H2: I ndividuals who finish a degree at a public college or university will realize a higher economic return than those who finish a degree at a non profit or for profit college or university. RQ3: How do the characteristics of individual near completers affect the likelihood that they will return to a college or university and complete a degree? H3: Factors that are important in predicting educational successes for traditional students will also be important in predicting w hether near completers finish degrees.

PAGE 57

46 C ontributions to the F ield Drawing on the limitation s identified in that literature review regarding research specific to near completers, t his section identifies ways in which evaluating these hypotheses will contribute to the field both empirically and theoretically. Theoretical development: Human capital and signaling t heories Although there does not appear to be any extant rigorous resear ch specifically focusing on the research question s posed above one could argue that human capital and signaling theories suggest there will be a positive individual economic return for near completers who finish degrees However, several characteristics of near completers make it difficult to reach that conclusion b ased on these theoretical frameworks. First, near completers are older than traditional students that formed the basis of the original theories and thus have a shorter window in whic h to recoup costs of education. Second, when these potential students retu rn and finish a degree, they are receiving a relatively limited amount of additional education, depending on how close to earning a degree they are. Empirical tests of the benefits of shorter term training programs for adults, though not completely analogo us to completing a college degree, have shown mixed results. Third, much research in these areas has focused on fairly idealized cases that may approximate reality for traditional aged students, but the underlying assumptions on direct and indirect costs m ay have more significant impacts on older students. Research on th ese question s could help to address whether changes in an n capital stock to include education, experience, and training, then it could be that the additional education later in life makes a smaller

PAGE 58

47 contribution to the overall stock in relative terms, resulting in a smaller impact of increases in education levels in the adult years. With near completers already possessing significant prior college credit, it could be that the additional postsecondary schooling does not offer significant economic benefits. Alternatively, completing a postsecondary degree could be a n important credential with tangible workplace benefits. Signaling theory and the sheepskin effect broadly speaking, would suggest that near completers who finish a degree should receive an economic benefit. However, i t could be that for near completers w ho have been in the workforce for many years, work experience supersedes the signal sent by their new diploma. In essence, their employers could already have a fairly complete view of the difficult to observe characteristics that a diploma represents as ex per ience sends a stronger signal than a new degree. Alternatively, it could be that earning a college diploma sends a new signal to employers work, it could be that an individual miscalculated the amount of education he or she needed to adequately match his or her characteristics, leading to underemployment. have changed over time perhaps through general maturity or due to life expe rience it could be that the new credential better captures these new factors and sends a more accurate signal to employers. This research effort will extend s these theo ries to address near completers by answering these different questions. Outcomes of pu blic and private o rganizations This research will also make a modest contribution to the stream of literature focused on whether management type impacts outcomes. T

PAGE 59

48 higher education wi ll limit that contribution to the subset of studies that use ownership and/or political control as the operational definition. Within that subset, there is limited research that examines whether management type may affect outcomes differently for subpopula tions such as adults. Within research on higher education, there are no apparent studies that examine outcomes for near completers by the type of management. Factors that affect the pursuit of additional education. Similarly, there is no apparent research about the individual characteristics that may be associated with the decision by near completers to return to postsecondary education to complete a degree. There has been some market research testing which messages resonate with near completers, but no ac ademic research has examined the characteristics of those who return to finish degrees compared to those who do not. This examination could help better target policies and programs while also filling in gaps in research about decisions of non traditional s tudents to pursue additional education. Empirical c ontributions The major empirical contribution that this research make s to the field is relatively straightforward. There is little rigorous research specific to whether near completers receive an individ ual economic benefit from returning to finish a baccalaureate degree beyond program evaluations that offer little comparison and few control variables. Policymakers (and potentially returning students) currently rely on simple and straightforward bivariate holders and those with some college, but no degree to make the case for broad programs aimed at these potential students. Beyond this central contribution, there are numerous other empirical questions that will be answered in the course of addressing the central question As a beginning

PAGE 60

49 p oint, there is little basic research on the character census category captures all those who have completed any colle ge credits and data based on this category underpins most of the policy and programmatic conclusions about near completers. Policy arguments to start degree completion programs serving near completers have focused on the fact that 22 percent of the adult population falls into this category ( U.S. Census Bureau, 2011 ). However, state programs are focused on a subset of this population those close to degrees and little is known about this group One side benefit of this research is estimates of the demogr aphic makeup of near completers. Additionally, little is known about the demographics and personal characteristics of this group. Although it would be relatively straightforward to conduct demographic research on the broader category from the census, it is not certain that this would accurately reflect the makeup of near completers. Given what is known about college completion rates by race and gender, it could be that this racial/ethnic background and gender are not evenly distributed in this population S uch information would be useful to practitioners seeking to reach and reengage this group, to policymakers seeking to address racial/ethnic differences in degree attainment, and to researchers examining differential returns to education. Additionally, it c ould be that the credit distribution is correlated with age. If near completers turn out to be much older than traditional students, or those who finish degrees do so after many years away from postsecondary education it could complicate policy arguments that serving these individuals will address future workforce needs because these individuals may not work for as many additional years.

PAGE 61

50 C osts for near c ompleters A general critique of literature on the benefits of education is that much of it focuses only on the wage premiums associated with greater attainment while ignoring costs. This is partly due to the assumptions ingrained in r, even when accounting for more realistic conditions, such as costs of tuition, most estimates of the wage premium for college degrees would suggest a positive return on investment even with extremely conservative estimates of future earnings and overesti mates of costs (Hout, 2012). Estimates of the economic return on college degrees almost uniformly show that the benefits substantially outweigh the costs with some notable differences by racial/ethnic subgroup or gender (Day & Newburger, 2002; Hout, 201 2; Julian & Kominski 2011 ; Kane & Rouse, 1995 ). Research has not specifically focused on the cost and potential returns for near completers who have a shorter window in which to recoup education costs. Compounding this, if near completers have to forego ear nings to pursue additional education, their indirect costs may be much higher than has been estimated for traditional students because the y likely have much higher salaries than traditional aged students although this would be offset if their subsequent w ages are also commensurately la rge. Estimating foregone earnings for near completers is an important addition to the field. As noted earlier, some r esearch on adults who earn degrees and certificates finds lower pre treatment earnings but little research focuses on whether those who do pursue additional education and training sacrifice wages to do so ( Ashenfelter, 1978; Blanden, et al., 2012; Jepsen & Montgomery, 2012; Marcus, 1986; Light, 1996 ). Conclusions based on this evidence differ, with some arguin g that lower incomes for adults who enter

PAGE 62

51 training and education programs means that opportunity costs and foregone earnings are barriers to adults pursuing further education, and others arguing that those with lower incomes may have greater economic needs and thus more reasons to enter school or a training program (Blanden, et al., 2012; Jepsen & Montgomery, 2012; Marcus, 1986). Further research will help identify how earnings change prior to and during additional education. One might also expect that esp ecially with the flexible adult focused programs being offered by institutions of higher education, f ew adults who are already employed and have the normal financial obligations of life would be likely to quit their jo bs and embark on a traditional educati on program full time with classes during the day, so their economic indirect costs may be less than those assumed by the standard model proposed by Becker. Alternatively, it could be that the growth in on line learning and other more flexible options is too recent to be captured by the dataset used here or that any commitment to complete a degree requires the student to forego some work opportunities 8 Measu feasible even with information about the school that he or she attended, due to financial aid packages and tuition discounting that affect the actual price paid. The analyses used in this research do not account for grants, scholarships, and tuition discounts, mainly 8 A true accounting of indirect costs includes not just foregone wages, but the cost of other non economic spending less time with family, and the p sychic costs associated with pursuing additional education. This study focuses exclusively on economic indirect costs. For additional discussion of non economic indirect costs see Heckman, Lochner, and Todd (2006).

PAGE 63

52 because such data are not available but given an appropriate data source this would be a valuable area for additional research. 9 Contributions to the f ield: Conclusion T his research will make a significant contribution to human capital and signaling theories, while also adding to literature on the differences between public and private organizations. Additionally, this research will contribute to ongoing policy discussion s about future workforce needs and degree completion programs. Although the policy community already explicitly assumes certain results that this effort purports to provide, as I have shown through this literature review, many of the assumed results are no t yet supported by solid empirical results. This research will provide important theoretical development and empirical evidence to either support or alter existing national, state, and local policies, while also providing some clarity about differences in performance of colleges and un iversities by management type. 9 Tuition discounting the phenomenon in which schools do not charge full price likely has a significant effect on estimated costs. Abel and Dietz (2014) provide additional discussion in the context of estimating returns to education.

PAGE 64

53 CHAPTER III DATA AND MEASUREMENTS This section describes the data source employed to test hypotheses proposed above. I present a wide range of descriptive data and illustrative figures that inform the chosen methodology, choice of variables, and approaches for producing estimates to test my hypotheses. Most of the descriptive data tables employ survey weights to present a representative picture of near completers. Data S ource To test the hypotheses posed above, I use the National Longitudinal Survey of Y outh 1979 (NLSY ). This longitudinal survey of 12,686 individuals has been conducted annually from 1979 to 1994 and biann ually from 1994 2012. All survey respondents were between ages 14 and 22 when the survey commenced in 1979. The survey is designed to be nationally representative, but is composed of three sub samples: the first is 6,111 nationally representative individua ls ; the second subsample is an oversample of 5,295 black, Hispanic, and low income non black, non Hispanic individuals; the final subsample is a group of 1,280 individuals enlisted in the military at the time of the initial survey (B ureau of Labor Statisti cs, n.d. ). Sample description. Within th e overall NLSY sample, I use 105 9 individuals operationalize this term as an individual who finished at least half of a baccalaureate d egree without completing it but then has at least one survey round in which he or she

PAGE 65

54 was not enrolled in college. 10 As this is the central definition for this study, additional justification and analysis of this operationalization is warranted. Due to the limited extant research on this topic, there is not a widely accepted operational definition to employ. Shapiro, et al. (2014), use two years of enrollment as marking significant progress toward a degree in delineating a grou p t but this is more of a g eneral report examining the number of students meeting these criteria Of the other studies purporting to examine returning students, there are few usable operational definitions. Hoyt and Allred ( 2008) evaluate a program that targeted students who had completed 30 credits (approximately one full time year of enrollment) who had not been enrolled in two years, while other research is not specific about the definition. The varied programs targeting these students that are currently in operation employ a range of definitions as well. A list of definitions from such programs is included in Table 5 though it is far from comprehensive. Although there is not a consensus, one key part of the definitions u sed is non enrollment. There is much variation in the cut off for the number of credits that marks substantial progress toward a degree. The definitions vary from any credits at all to 75 percent of the way toward a degree Using 50 percent of the credits necessary for a degree represents middle ground among these varied definitions. 10 This measure is based on years of postsecondary completed An individual who attended college for two years part time but only earned credits necessary to finish one quarter of a degree would have a high grade completed of 13 even though he or she completed two calendar years in postsecondary education.

PAGE 66

55 Table 5 : College and state definitions for degree completion programs State Definition Source Arkansas 75 percent of credits necessary for a degree, at least 22 years old, out of college at least two years. Lane, Michelau, & Palmer (2012) Brigham Young University At least 30 credits (25 percent of those necessary for a degree), not enrolled for two years. Hoyt & Allred (2008) Colorado 75 percent of credits necessary for a degree, at least 25 years old, no longer enrolled. Lane, Michelau, & Palmer (2012) Connecticut At least 12 credits completed. At least 18 months of non enrollment. Connecticut House Bill 5050 (2014) Kentucky 2/3 of the credits necessary for a degree, not enrolled. Kentucky Council on Postsecondary Education (2008) Maryland At least 90 credits competed, not enrolled. Maryland Senate Bill 0740 (2013) Minnesota Some college credits (not specified) but no degree, not enrolled. Minnesota State Colleges and Universities (2015) Nevada No credit limit, has not been enrolled in previous year. Lane, Michelau, & Palmer (2012) New Jersey Credit definition left up to schools, not enrolled Lane, Michelau, & Palm er (2012) North Dakota Completed 70 percent of credits, not enrolled Lane, Michelau, & Palmer (2012) Oklahoma 72 credits, at least 21 years old, not enrolled, completed general education requirements. Oklahoma State Regents for Higher Education (n.d.) South Dakota 90 or more credits, not enrolled Lane, Michelau, & Palmer (2012)

PAGE 67

56 Other operational definitions are plausible, as well. I consider the following alternatives 1. Individuals who complete at least 50 percent of a baccalaureate degree and then make no further progress on their degree for thr ee years The goal of the second part of the definition is to ensure that the individual is no longer enrolled in postsecondary education. While this slow progress could indicate that, it could also be that some of these individuals are progressing slowly while enrolled, likely due to being part time students. Given that individuals report their education status during each round, for the years in which the survey takes place every two years, it is not possib le to determine a three year gap with precision. Thus, the operational definition is that an individual remains at the same level of education for three consecutive survey rounds for 1979 1994 and two consecutive survey rounds from 1994 2012. The latter y ears effectively become a pause in progression of at least four years. 2. Individuals who complete at least 50 percent of a ba ccalaureate degree and have a spell of non enrollment that lasts at least two years. This definition accomplishes the main goal of en suring that an individual is no longer enrolled, but similar to previous definition, it is not clear how this definition improves on the preferred one. The only difference is in the length of time away from earning a degree Considering only the years 1994 2012, this results in an identical population as the main definition due to the biennial nature of the survey during these later years. For the years 1979 1994, it eliminates from consideration those individuals who were not enrolled for only one survey r ound.

PAGE 68

57 3. Individuals who complete at least 75 percent of a baccalaureate degree and then report a spell of non enrollment. From a policy perspective, if the goal is to identify those individuals who have completed significant credits and can complete degrees relatively quickly, selecting a lower cut off does not seem appropriate as those who are less than 50 percent of the way towards a degree would still be technically in lower division classes The descriptive data presented below, as well as the results an d discussion that follow rely on the operational definition chosen above. I include sensitivity tests comparing results using the potential alternative definitions and a further discussion of the imp lications for this operational decision in the appendix Des c riptive data of the background characteristics of these individuals who attained near completer status as defined above, are presented in Table 6 The data show that the average age was just under 18 when individuals in the sample were first interview ed. Parents of the sample respondents completed between 12 and 13 years of for mal schooling on average, and 12 percent of the sample fell below poverty line in Vocational Apt itude Battery (ASVAB), which is a standardized test for qualificat ion for the armed forces frequently used in literature on education and earnings as a control for ability (Black & Smith, 2004; Monks, 2000; Kane & Rouse, 1995; Tanigu ichi & Kaufman, 2005). S urvey respondents took the ASVAB at the outset of the survey and their ages ranged from 14 22 when taking the exam, which would bias the scores upwards for older individuals and downwards for younger individuals. I adjust these scores rescaling them based

PAGE 69

58 Table 6 : Descriptive statistics for those who attained near completer status Variable Mean Linearized Std. Err. Min Max 1979 Age 17. 72 0.10 14 22 Magazine s in the home (1979) 0.76 0.02 0 1 Newspapers in the home (1979) 0.88 0.01 0 1 Possessed library card (1979) 0.83 0.01 0 1 12.45 0.10 0 20 12.84 0.15 0 20 Fam ily below poverty line (1979) 0.12 0.01 0 1 ASVAB Score (1979) 68.08 0.82 3 99 Male 0.48 0. 02 0 1 Race/Ethnicity Hispanic 0.06 0.01 0 1 African American 0.16 0.01 0 1 Non African American/Hispanic 0.77 0.01 0 1 N 1059 Note: Data are weighted using sampling weights provided by NLSY with errors adjusted to account for stratified cluster sampling strategy. procedure follows Monks (2000) and Taniguichi and Kaufman (2005). A small portion of the sample (< 4%) did not report ASVAB scores. I follow Light and Strayer (2004) and assign these individuals the mean score. 11 The weighted average for those individuals in the subsample is just above the 65 th percentile. Just under half of the subsample is male, and th e race/ethnicity backgrounds of near completers is 77 percent non African American/Hispanic, 16 percent African American, and 7 percent Hispanic. Survey attrition and changes Although the survey is widely used and accepted in social science research, th ere have been a number of changes in the NLSY79 over time that could potentially affect findings and must be acknowledged. First, as in all longitudinal surveys, non random attrition is a concern. Between 1979 and 1994, the 11 Dropping those individuals with missing ASVAB scores from the analyses presented later in this paper does not subs tantively affect results when compared to including them in the analyses that rely on ASVAB scores.

PAGE 70

59 overall retention rate was 89 pe rcent and 75 percent of the individuals from the main sample who started the survey in 1979 responded in 2010 (B ureau of Labor Statistics, n.d. ). T he overall attrition rate is relatively small compared to other longitudinal surveys over such a time period (Hernandez, 1999 ) More recent analysis shows that the NLSY reinterview rate averages about 96 percent, which compares favorably to other large longitudinal surveys (Schoeni, Stafford, McGonagle, & Andreski, 2013). There is li mited evidence of non random attrition as shown in Table 7 which shows unweighted descriptive data for the sample used in 1979 compared to the portion of that sample that was also interviewed in 2012. The most striking difference between the 1979 and 2012 samples is t he decrease in the proportion of the sample that was in poverty in 1979. While a portion of this may be due to standard attrition, it is likely that the bulk of this change comes from the decision by survey administrators to drop par t of the low income ove rsample i n 1990 (Bureau of Labor Statistics, n.d.). R esearchers have examined attrition in the NLSY and conclude that the relatively small differences are not likely to bias estimates (Sen, 2006) Further, NLSY 79 is widely used for research in a variety o f su bject areas and findi ngs based on the data are generally accepted in academic literature (MaCurdy, Mroz, & Gritz, 1998). Still, it is possible to adjust the data using survey weights provided by NLSY to correct for attrition in addition to correcting f or oversampling. Repeating the descriptive data with appropriate survey weights applied to the subsample s of interest minimizes the differences due to attrition, as can be seen in Table 8 A djusted Wald test s show that there are no significant differences in the means for any of the variables is this table

PAGE 71

60 Table 7 : Changes in descriptive data (unweighted) 1 979 to 2012 1979 2012 Variable Mean SD Mean SD 1979 Age* 17.9 1 2.35 17.63 2.24 Magazine s in the home (1979) 0.68 0.47 0.66 0.48 Newspapers in the home (1979) 0.82 0.39 0.79 0.41 Possessed library card (1979) 0.79 0.41 0.78 0.41 11.82 3.16 11.73 3.16 12.05 3.95 11.86 3.96 Fam ily below poverty line (1979)* 0.21 0.41 0.20 0.40 ASVAB Score (1979) 61.86 24.79 59.33 25.66 Male 0.46 0.50 0.44 0.50 Race/Ethnicity Hispanic 0.16 0.37 0.18 0.38 African American 0.29 0.46 0.36 0.48 Non African American/Hispanic 0.54 0.50 0.46 0.50 N 1059 648 *Denotes a statistically significant difference in means at p<.05. Table 8 : Weighted descriptive data, 1979 and 2012 1979 2012 Variable Mean Linearized Std. Err. Mean Linearized Std. Err. 1979 Age 17. 72 0.10 17.8 5 0.1 1 Magazine s in the home (1979) 0.76 0.02 0.7 5 0.02 Newspapers in the home (1979) 0.88 0.01 0.8 7 0.0 1 Possessed library card (1979) 0.83 0.01 0.8 2 0.02 12.45 0.10 12. 35 0.1 3 12.84 0.15 12. 71 0.18 Fam ily below poverty line (1979) 0.12 0.01 0 .13 0 .02 ASVAB Score (1979) 68.08 0.82 67.32 1.02 Male 0.48 0. 02 0.49 0.02 Race/Ethnicity Hispanic 0.06 0.01 0.07 0.01 African American 0.16 0.01 0.17 0.01 Non African American/Hispanic 0.77 0.01 0.77 0.01 N 1059 648

PAGE 72

61 Educational attainment v ariables. key variable in this analysis. operationalized as an individual who completed more than half of a baccalaureate degree without completion and had a period of non enrollment. degrees degree important for understanding the broader credential completion environment, they are beyond the scope of this study. Further research into these types of credentials, and their impact on individual wages and social outcomes such as filling workforce demand is certainly warranted. This research is intended to focus specifically on outcomes for near completers who finish baccalaur eate degrees, which could be confounded by including Within the subset of i ndividuals who attain the status of near completer at some time during the survey, respondents are divided into four categories in any year after they first attained near completer status: 1) near completers, operationalized as described above but who do not fall into one of the following three categories ; 2) near completers, pre enrollment, which includes those individuals who are near completers and who report enrolling in postsecondary education in the subsequent interview period; 3) near completers enrolled, which includes those individuals who are near completers who report being enrolled in college; finished degrees, which includes those individuals who were near completers at one point but returned and completed a bacca laureate degree.

PAGE 73

62 The second category is included to control for the Ashenfelter Dip where income declines just prior to an individual deciding to pursue additional education or training The third category has two purposes. First, including this variable controls for a temporary drop in income caused by foregone wages that could bias estimates of the treatment effect in the same way as the Ashenfelter Dip. Second, estimating this coefficient will provide important information about indirect costs for calc ulating the cost of attending college for near completers. Collapsing those four categories to just two for the purposes of providing descriptive data shows how the population of near completers and those with degrees changes over time. The changes over t ime in the number of near completers and those near completers who finished degrees are displayed in Figure 1 The number of near completers grows rapidly through 1987 at which point the growth starts to level off. The number of near completers who finish degrees continues to grow slowly but steadily through out the rest of the survey. T his initial analysis ignores attrition, and once an individual attains near completer status, he or she is counted in the categories above in perpetuity, regardless of whether he or she responded to the survey in a given year. Accounting for attrition shows slight differences, as can be seen in Figure 2.

PAGE 74

63 This second figure shows a substantial decline in the near completer population between 1989 and 1990, likely reflecting the changes to the sample and the decision by administrators to discontinue interviews with certain subsamples. Following that large drop, the number of near completers declines slightly through the rest of the survey. The number of near completers who finished degrees grows somewhat quickly through 1990 (though less rapidly than the population of near completers), before its growth rate decreases slightly, growing slowly through the rest of the survey. Some of this growth Figure 2 : N ear completers and near completers finishing degrees over time, accounting for attrition Figure 1 : N ear completers and near completers finishing degrees over time

PAGE 75

64 likely contributes to the declines in the population of near complete rs who did not finish degrees. Which near completers finish degr ees? One key question for this research effort is about the individual characteristics that may predict whether a near completer returns to finish a degree or not. Comparing those who attained the status of near completer but never finished a degree with those near completers who at some point finished a baccalaureate degree shows notable statistically significant differences as can be seen in T able 9 Several characteristics that are associated with larger educational attainment Table 9 : Demographic data by educational attainment Near Completers, Never Finished Near Completers, Finished Degrees Variable Mean Linearized Std. Err Mean Linearized Std. Err 1979 Age 17. 75 0.12 17.67 0.17 Magazine s in the home (1979) ** 0.72 0.02 0.82 0.02 Newspapers in the home (1979) ** 0.86 0.01 0.91 0.02 Possessed library card (1979) *** 0.79 0.02 0.89 0.02 *** 12.06 0.12 13.14 0.18 *** 12.25 0.17 13.82 0.25 Fam ily below poverty line (1979) *** 0.14 0.01 0.08 0.02 ASVAB Score (1979) *** 63.72 1.02 75.71 1.26 Male 0.48 0.02 0.48 0.03 Race/Ethnicity**** Hispanic 0.08 0.01 0.03 0.01 African American 0.19 0.01 0.12 0.01 Non African American/Hispanic 0.73 0.02 0.85 0.01 N 740 319 *Denotes a statistically signifi cant difference in means at p<.10 ** Denotes a statistically signifi cant difference in means at p<.05 ***Denotes a statistically significant difference in means at p<.01 ****Denotes a statistically significant difference in distribution at p<.01 using a corrected Data are weighted using survey weights provided by NLSY.

PAGE 76

65 elsewhere in the literature show significant and substantively large differences, including paren tal education receipt of magazines and newspapers, possession of a library card and ASVAB scores which all tend to be higher among those who finish degrees Initial poverty levels tend to be lower among degree finishers. The racial and ethnic distribution of the two groups also appears to be different, with a higher percentage of individuals who are neither His panic nor African American among degree finishers. The difference in the distributions is statistically significant. This initial analysis is suggestive of some important differences between those who finish degrees and those who do not, but estimating the impact of different variables on whether an individual completes a degree will require an event history analysis. Th is methodology will be described in detail below, but essentially this approach can examine the event of interest ( returning to complete a baccalaureate degree in this case) over time and estimate the impact of different time varying and time invariant ch aracteristics on the likelihood that an individual will earn a degree. This methodology also lends itself to some useful descriptive statistics and graphics presented below. The initial descriptive analysis, shown in Figure 3 is a Kaplan Meier survival es event of interest has not occurred and the individual remains a near completer rather than a degree holder 12 The X axis shows th e length of time in years and the Y axis estimates the proport ion of the population that remain s near completers rather than becoming near completers who graduated Because survival is 12 One of the classic uses of event history analysis is the examination of the effect of medical treatments on

PAGE 77

66 Figure 3 : Survival estimates for near completers equated w ith not graduating, the farther the curve drops below one, the more near completers that have graduated. This figure shows a curve that flattens out over time, suggesting that most near completers who finish degrees do so within the first 10 years of attai ning that status. Additionally, the curve is even steeper in the first four years, suggesting that the near completers are less likely to finish degrees as time passes Part of this is likely due to how college credits for reenrolling students are treated as former students try to reenroll Those who transfer to new schools tend to lose credits as their new school may not accept all of the previously earned credits, or may accept them for elective credit only, forcing the transfer student to re take classes that count towards a major contributing to lower graduation rates (Government Accountability Office, 2005). Additionally, some schools may have policies affecting whether credits are accepted depending on how old they are. Although schools ma y agree to accept all credits, regardless of age, it is often up to individual departments to determine whether older credits are accepted for major credit, or electives again forcing returning students to retake classes they have already

PAGE 78

67 completed. There is little research or policy analysis of this treatment of credits, but it is an important consideration for returning students who may be set back several semesters by enrollment policies at the schools they attend. This Kaplan Meier analysis can also be disaggregated by key variabl es. Figure 4 shows the same survival curve disaggregated by gender The figure shows slight differences in the survival curves for men and women, with men slightly more likely to earn degrees in the first 15 years of being a ne ar completer, while women overtake men later in the study period. However, the differences between the two estimates are not statistically significant using a log rank test. Figure 4 : Survival estimates for near completers by gender Figure 5 shows a similar analysis for different racial/eth nic groups. These estimates clearly show different survival rates for near completers of different racial/ethnic backgrounds, with Hispanic near completers appearing less likely to finish

PAGE 79

68 Figure 5 : Near completer survival rates by racial/ethnic group degrees than African Americans and individuals of other racial/ethnic backgrounds. The difference between the latter two groups is more complicated with African Amer icans falling behind those from other racial/ ethnic backgrounds initially, but then closing the gap over time, which suggests there may be differences about when individuals from different racial/ethnic backgrounds return to finish degrees. A log rank test shows that the differences between these survival curves are statistically significant. These results are suggestive that there may be differences in the likelihoods for near completers to finish degrees by racial/ethnic background. I also examine surviva Meier estimates are presented below in Figure 6 This shows that those near completers whose families were in poverty in 1979 are less likely to finish degrees over the course of the survey. The diff erences between the two curves are statistically significant at p<.01.

PAGE 80

69 Figure 6 : Near completer survival rates by familial poverty status Additionally, across all of these Kaplan Meier estimates, it is clear that the rate at wh ich near completers finish degrees changes over the length of time since they have attai ned that status. Income and degree completion. While there are clearly changes within the sample in the education status levels of individuals, one important question for this a nalysis is whether degree completions lead to changes in income levels. I ncome is clearly dependent on numerous factors beyond just education al attainment, but examining average incomes of near completers compared to near completers who finished degrees can still be instructive The Table 10 examines income differences between the two cate gories at three points in time. These data show that those near completers who finis hed degrees have earned higher wages with strong statistical significance for all three years considered. Additionally, the gap appears to widen over time, with the premium being approximately one third of non finisher wages in 1992 rising to slightly more than have of non finisher wages by 2012. Given the large number of other variables that may also impact income

PAGE 81

70 (and could be correlated with education status), a more detailed multivariate analysis is required. Table 10 : I ncome by educational attainment 1992, 2002, & 2012 Near Completers, No Degree Near Completers, Finished Degrees Year Mean Income Linearized Std. Err. N Mean Income Linearized Std. Err. N 1992 ** $36 160 1,600 414 $49,914 4,821 139 2002 *** $5 1,008 2,888 381 $74, 625 6,385 169 2012** $55, 0 41 3,972 318 $ 84, 858 6,89 3 197 Note: All income adjusted to 2012 dollars. Only respondents with non zero income reported. ** Denotes a st atistically significant difference in means at p<.01 Sampling weights and adjustments for stratified cluster sampling applied. Foregone wages and i ndirect costs. A key issue for this study is the cost borne by a near completer who returns to school in the form of foregone wages. H aving likely worked for more years than t raditional aged students, they may earn higher wages and could thus face substantially higher foregone wages if they are not able to work and complete s chool at the same time. Table 11 shows mean earnings for near completers compared to near completers who are enrolled in college and near completers who graduated at three points in time. Enrollment is a self reported measure and could include both full and part time students. These data show that near completers who are enrolled have substantially lower ea rnings (with statistical significance) in two of the three years Table 11 : Income for near completers, by enrollment and graduation status. Near Completers, Not Enrolled Near Completers, Enrolled Near Completers, Finished Degrees Year Mean Income Linearized Std. Err. N Mean Income Linearized Std. Err. N Mean Income Linearized Std. Err. N 1992 *** $38, 054 1,8 64 338 $27,51 7 2,1 24 76 $49 914 4,821 139 2002 $51, 537 3,0 07 351 $4 3,551 10, 153 30 $74, 625 6, 385 169 2012 *** $56, 241 4,111 299 $26, 883 5, 554 19 $84,8 58 6,893 197 Note: All income adjusted to 2012 dollars. Only respondents with non zero income reported. *** Statistically significant difference in means at p<.01 for non enrollees and e nrollees only. Sampling weights and adjustments for stratified cluster sampling applied.

PAGE 82

71 compared to near completers who are not enrolled T he se results are suggestive that there may be some indirect costs associated with returning to finish a degree even for students who are part time The number of enrolled students in a given year (ranging from 76 down to 19 in 2012) are relatively small, so caution is warranted b efore reaching conclusions. The A shenfelter D ip. The panel nature of the datas et makes it possible to control for the potential existence of the Ashenfelter Dip by examining incomes in the survey round prior to reenrollment. For survey rounds prior to 1994, this represents the year prior to reenrollment. For survey rounds after this p eriod, reported income is for two years prior to reenrollment due to the biennial nature of the survey from 1994 through 2012. 13 The descriptive data presented in Table 1 2 compares incomes of near completers with near completers who enroll in the following survey round. Table 12 : Pre enrollment income for returning near completers Attainment Status Mean Income Linearized Std. Err. Observations Near Completers a $40,510 583 7230 Pre enrollment a $32,743 2,191 3 50 Note: All income adjusted to 2012 dollars. Only respondents with non zero income reported. a Statistically significant difference in means at p<.01 for near comp leters and preenrollees Sampling weights and adjustments for stratified cluster sampling applied. Instead of only examining individual years, I consider average salaries for all observations of near completers and all observations where the individual is in the pre enrollment category. The difference shows strong statistical significance, and suggests that there may be a decline in income prior to reenrollment that could bias estimations if it is not accounted for in the final model. 13 Although th is presents potential measurement issues, sensitivity testing of the regression models presented later shows no difference if the sample is restricted to 1994 so that the income variable is truly the year before enrollment, or the full sample, with biennia l responses from 1994 2012.

PAGE 83

72 Direct costs of college attendance. As described in greater detail in the following chapter, accounting for the direct costs borne by individuals returning to complete degrees is necessary for estimating the actual rate of return. As opposed to foregone wages, where near completers are like ly to sacrifice more money than traditional students, direct costs as a percentage of their earnings are likely to be lower than those borne by traditional students. With the available data, however, it is not possible to determine the actual tuition paid by any individual near completer returning to school. T his presents difficulties for estimating an actual rat e of return. Instead, as described in greater detail in th e section on methodology, I present a range of scenarios with different levels of direct In developing the range of direct costs to use, I consider differences in direct costs by sector. T able 1 3 shows average tuition rates by management type over the course of the survey. While this is only a rough estimate of what an individual might have paid for attending additional postsecondary education, it does, in conjunction with estimates of foregone wages, provide a basis for approximating actual rates of return, as described in greater detail in the following chapter. One important observation from this table is that tuition prices have increased over time, even after adjusting for inflation. This likely means that estimates of the return to degree completion will depend to some extent on the year in which an individual completes his or her degree. If the return to degree completion remains constant, t hose who completed degrees later will bear higher costs and thus likely receive a smaller economic return relative to those who completed earlier In creasing returns to degree completion over time may offset this tuition growth, however (Athreya & Eberly, 2015)

PAGE 84

73 Table 13 : Average tuition and fees by school management type, 1980 2012 Year Public Tuition Private Tuition Non profit For profit 1980 $2,111 $9,501 1981 $2,199 $9,947 1982 $ 2,391 $ 10,292 1983 $ 2,566 $ 10,845 1984 $ 2,642 $ 11,435 1985 $ 2,755 $ 12,106 1986 $ 2,892 $ 12,921 1987 $ 3,020 $ 13,728 1988 $ 3,091 $ 14,010 1989 $ 3,189 $ 14,601 1990 $ 3,208 $ 14,907 1991 $ 3,486 $ 15,509 1992 $ 3,751 $ 15,873 1993 $ 3,948 $ 16,453 1994 $ 4,056 $ 16,810 1996 $ 4,278 $ 17,898 1998 $ 4,465 $ 18,571 2000 $ 4,550 $ 20,270 $ 13,247 2002 $ 5,056 $ 21,692 $ 14,133 2004 $ 5,967 $ 23,147 $ 15,475 2006 $ 6,316 $ 24,348 $ 15,916 2008 $ 6,691 $ 25,975 $ 15,169 2010 $ 7,345 $ 27,156 $ 14,670 2012 ** $7 904 $28,533 $14,955 All figures presented in 2012 dollars. *NCES did not calculate private tuition separately for private non profit and private for profit colleges and universities until 1999. **2012 Figures not available. Average tuition calculated by adding the averaging change in tuition from 2000 2010. Source: National Center for Education Statistics (2013). Public vs. private graduates. Finally, descriptive data on near completers who finished at public colleges and universities are compared to those who finished at non profit and for profit colleges and universities in Table 1 4 The sector of college or university from which an individu al graduated is derived from the NLSY restricted use

PAGE 85

74 Table 14 : Descriptive data for near completer s who finish a degree, by sector Variables Public Private, non profit F or profit Mean Linearized Std. Err. Mean Linearized Std. Err. Mean Linearized Std. Err. Mean Income (20 12 ) $ 91,264 9, 401 $ 75,816 13, 730 $70,001 15,6 44 Yrs. since graduation (2012) a 12.42 0.51 11.30 1.02 6.58 1.74 Mean Income (1 yr. post grad.) bc $ 28,712 29,916 $ 35,709 29,524 $49,574 18,708 Age (at graduation) bd 29.89 6.75 32.15 8.65 42.69 8.15 Magazines (1979) 0. 83 0.03 0.81 0.05 0.67 0.14 Newspapers (1979) e 0. 89 0.02 0.9 4 0.02 0.7 8 0.1 5 Library card (1979) 0. 89 0.02 0.89 0.04 0.69 0.16 1 3.01 0.25 13.32 0.33 13.17 1.19 13. 70 0. 34 13. 90 0. 47 13.06 1. 79 Family below poverty line (1979) 0.0 6 0.02 0.0 7 0.0 3 0.0 3 0.0 3 ASVAB Score 75 16 1 66 78 04 2 46 70 14 7 15 N 1 97 75 13 Note: All income figures adjusted to 2012 dollars. Only respondents with non zero income reported. Descriptive data on gender and racial/ethnic background suppressed due to small cell sizes. a Years since graduation shows a statistically significant difference between public and for profit at p<.01 and a statistically significa nt difference between private and for profit at p<.05. b Due to these events occurring in different years, unweighted means are reported for these two variables, along with traditional standard errors. c Post graduation income shows a s tatistically significant difference in means for public and for profit at p<.10. Receipt of newspapers shows statistically significant d Ages show statistically significant differences between for profit and both other categories at p<.01. Differences in age between pu blic and private are statistically significant at p<.10. Sampling weights and adjustments for stratified cluster sampling applied. e Receipt of newspapers shows statistically significant differences at p<.10 between public and non profit schools. datase t, following Monks (2000). The survey includes the Federal Interagency Committee on Education (FICE) codes to identify the school. Sector is then determined by the Carnegie c lassifications. The number of individuals who graduated from for profit schools is quite small, perhaps due to the fact that for profit institutions have only grown significantly in the last decade and the number of near completers who finished degrees has declined in recent years. While the number of near completers who finished degree s

PAGE 86

75 at for profit schools represents only 6 percent of the total number of finishers, for profit graduates account for 15 percent of those who have graduated since 2000. The descriptive data show that individuals who graduate from for profit schools may have higher incomes immediately after graduating (although the difference is not significant) while public school graduates have higher earnings in the final year of the survey The differences in earnings in the final year of the survey are statistically sig nif icant between public and non profit graduates only. There is a statistically significant difference between means for the age when individuals graduate, with for profit graduates tending to be older on average when they finish than graduates from other schools, which suggests that they are finishing their degrees much later in their careers. The only other variable with a statistically significant difference is the receipt of newspapers in the family home growing up, with non profit school graduates more likely to receive them than public school graduates. Likely due to the limited number of individuals who graduated from for profit schools none of the other variables show statistically significant differences. Data Limitations. As discussed above, non r andom attrition is a concern, although previous research suggests that the level of attrition here should not overly bias results. There are other limitations to the data that should be acknowledged. measure of racial/ethnic background is prob lematic with only three categories, and this may limit the generalizability for findings based on these sub groups. With these data, however, there are no alternative measures. As will be discussed in more detail in the chapter on methodology, I combine Af rican Americans and Hispanics into a single category to improve the size of the subsamples. While this does allow for analysis of

PAGE 87

76 important differences between predominantly White individuals (although technically, individuals of other racial/ethnic backgr ounds could be included), it may mask important differences between Hispanics and African Americans. Additionally, key variables such as education level and income are self reported which could be a concern (Ashenfelter & Krueger, 1994) Should measuremen t error for these variables be random, it would not bias estimates but only increase standard errors ( Wooldridge 2006). However, if individuals systematically overstate or understate their earnings, education level, or other key variables, it could bias e stimates ( Bound & Krueger, 1991 ; Wooldridge 2006). In the particular models pr esented in the following chapter where income is a dependent any of the independent vari ables used (Bound & Krueger, 1991). A study comparing both self reported income for a longitudinal survey and income reported to the Social Security significant differences between the two they are substantively small and should not overly bias estimates (Bound & Krueger, 1991). Self reported education is also subject to non random error. Ashenfelter and Krueger (1994) find that about ten percent of the variance in self reported schooling levels (when years of schooling are reported) is due to error. Other research suggests that individuals are much more accurate in reporting degree levels attained than in reporting years of schooling (Kane, Rouse, & Straiger, 199 9). While this does raise some concern for the estimates I present below, my estimations do not attempt to examine the marginal return to a year of postsecondary education which minimize bias. Further, the decision by NLSY administrators to switch to comp uter assisted interviews has been shown to

PAGE 88

77 reduce errors and improve accuracy in self reported data ( Baker, Bradburn, & Johnson, 1995; Tourangeau & Smith, 1996). The data used in this analysis only consider the type of school from which a near completer gr aduates. Many returning students may reenroll multiple times before starting the enrollment spell that leads to degree completion. More granular analysis of this process would likely benefit policymakers; however, it is beyond the scope of this study. Furt her the analysis of outcomes by sector is somewhat limited by the uneven distribut ion of graduates in each sector, especially the small number of graduates from for profit schools. Data: Conclusion In spite of these potential limitat ions, the NLSY dataset offers substantial depth of information and its continuity since 1979 provides a rich longitudinal cohort. Descriptive data and initial analyses are suggestive that near completers who finish degrees may receive substantial wage premiums; however, it coul d also be that they forego wages in pursuit of these degrees. These initial analyses suggest that models will have to control not just for characteristics like race, gender, and family background, but also an enrollment. Initial analyses of the outcomes for graduates of different sectors show suggestive differences in incomes between sectors, as well as differences in the age at which individuals graduate, but further investigation is warranted.

PAGE 89

7 8 CHAPTER IV METHODOLOGY To address the research questions posed at the outset of this dissertation I employ two series of models. The first series is designed to estimate the income benefit from completing a degree including estimates for the income benefit from gra duating from public, private, and for profit colleges and universities. While these models provide estimates of the wage premium for completing a degree, this does not represent the actual e conomic return, except under a set of assumptions that are unreali s tic for near completers. The following section of this chapter explains how I use estimates of the wage premium, along with other data, to determine the wage, tuition, and work life conditions under which a near completer who finishes a degree is likely t o earn a positive economic return. This chapter concludes with a section describing the event history analysis model used to determine how certain observable characteristics affect the likelihood that an individual will return to finish his or her degree. Individual Level Fixed E ffects As noted in the review of literature above, ability bias is one of the key sources of endogeneity in research on the economic return to education. The concern is that unobservable characteristics, such as ability or motivatio n, may cause near completers to return to finish a degree, while also causing them to earn higher wages independent of their education levels. The models and methodological approach here seek to control, to the greatest extent possible the s e and other sou rces of bias that could impact findings. The model s in this series attempt to estimate changes in income that are due to an individual near completer returning to finish his or her degree. Although the dataset

PAGE 90

79 offers a rich collection of control variables there is still concern about omitted variables and the endogeneity of education, particularly if the ASVAB score does not fully capture innate ability. I take advantage of the longitudinal nature of the data to employ an individual level fixed effects estimation which and post treatment earnings, and in doing so, controls for all time invariant characteristics of individuals that could bias results ( Allison, 2009; Angrist & Pisch ke, 2009; Cellini & Chaudhary, 2014 ; Jepsen, Troske, & time invariant, this approach will eliminate the potential for ability bias. Additionally, this approach controls for observable characteristics that do not vary over the time period in question, such as gender, racial/ethnic background, parental education, high school The model i s specified as follows: Y i t = i 1 d it + 2 x it + i + t + it where Y is the logged annual earnings (adjusted for inflation) for individual i in year t The key education variable is d which is a series of dichotomous variables for four potential education levels: near completer; near completer just prior to enrollment; near completer, enrolled; and near completer, finished degrees The education attainment variables are mutually exclusi ve in any given year. In the analyses that follow near completer is omitted and that category serves as the comparison to the other categories The additional categories help eliminate two potential sources of bias. The Ashenfelter Dip may lead to biased e stimates if incomes decline prior to reenrollment, and then naturally recovers. Similarly, if there are substantial foregone wages while near

PAGE 91

80 completers are enrolled in school and working to finish a degree, estimates for the degree completion coefficient would be biased as these individuals finish school and no longer lose earnings because they are enrolled. Additionally, the model includes x which is a vector of time variant characteristics of individual i in year t including age, experience, weeks out of the labor force, weeks unemployed, number of children, and health status Year dummy variables are included in while is the individual intercept, which includes the time invariant observable and unobservable characteristics of individual i such as gender, racial/ethnic background, innate ability, ASVAB scores, parental education, family socio economic status, high school academic success, and innate ability. is the individual error term at each year represents the combined error of all time invariant variables with relation to y (Allison, 2009) Following standard practice in return to education lit erature, a quadratic term for experience here a derived variable counting each year an individual works at least 26 weeks is also inc luded ( Becker, 1993 ; Mincer, 1974 ) However, some research shows that limiting the specification to a quadratic experience term introduces significant bias into the model, as the error term is correlated with experience (Murphy & Welch, 1990). Including on ly the quadratic term may understate income growth early in an ind career, and overstate the late career decline (Murphy & Welch, 1990). To correct for this potential bias, I include higher order experience terms. Because the individual level fi xed effects approach absorbs several variables that are of interest into the term that includes all time invariant characteristics, I repeat the model using interaction terms for racial/ethnic background and gender. This approach

PAGE 92

81 shows whether there are di fferences in the impact of certain characteristics depending on group membership and whether sector of graduation impacts the wage premium an individual receives. The individual level fixed effects approach essentially captures the trea tment effect of graduation by comparing pre and post degree earnings. This approach is not suitable for many analyses of the economic returns to education because for traditional students that proceed directly through the education pipeline, their actual earnings prior to earning a degree are likely quite low due to the types of employment available to full time students. However, with near completers, who have opportunities to work full time before treatment, this approach is appropriate and can control for major sources of omitted variable bias (Cellini & Chaudhary, 2014 ; Jepsen, Troske, and Coomes, 2014). To determine whether racial/ethnic background, gender, initial SES, and the sector of the school from which the individual graduates affects his or her earnings, I repeat the model as follows: Y i t = i 1 p it + 2 x it 3 i + t + it The only change from the initial model is replacing the education status term with p which is a series of interaction terms that reflect the same education categories as above, combined with these important time invariant characteristics. The interaction is applied to time periods prior to enrollment, enrollment itself, and post graduation. This will help determine whether there are differences in wage premiums and foregone wages by race/ethnicity, gender, initial SES, and sector Survey weights and regression. The question of whether to apply survey weights provided by NLSY to the data used in the regression analyses described above is

PAGE 93

82 complicated. Weights were applied to derive the univariate descriptive statistics presented above, which is necessary to present data that are representative of the population (Solon, Haider, & Wooldridge, 2015). Given that NLSY oversamples several population subgroups, it could be argued that weighting the data for the regression analysis would also be an appropriate course of action to provide more accurate estimates of the effect of degree completion on earnings. However, methodological research suggests that this may not be the best approach given the model proposed above. While applying survey weights to the dataset makes intuitive sense, several econometricians show that this can lead to inefficient estimates and errant conclusions Solon, Haider, and Wooldridge (2015) identify three typical justifications for weighting data: the first is correcting for heteroskedasticity due to differences in group sample size when the unit of analysis is not an individual ; the second involves trying to identify a population effect that is generalizable and representative; and the third is to correct for endogeneity of the sample, whereby the dependent variable is a function of the different criteria used to establish subpopulations in the sampling scheme. Winship and Radbill (1994) similarly identify errors with using sampl e weights in OLS regressions and argue that it should only be done when the sampling criteria are a function of the dependent variable (as in the third reason cited above). In those cases, if at all possible, they argue that the model should be respecified rather than use sampling weights, which can result in inefficient and biased estimates. Of these three reasons for using sampling weights, heteroskedasticity due to different group sizes is not relevant because I use individuals as the unit of analysis. 14 14 As discussed in the results, post regression diagnostics do show that other types of heteroskedasticity are present, but these are accounted for with clustered robust standard errors.

PAGE 94

83 Attempting to identify an average population effect could also be justification for weighting, particularly when heterogeneous effects may be present (Solon, Haider, & Wooldridge, 2015). However, the preferred approach (and the one that I take here) is to model the heterogeneous effects and report the differences rather than attempt to determine an average across the population (Solon, Haider, & Wooldridge, 2015). Finally, the endogeneity of the outcome variable (income) to the sampling criteria is account ed for by the fixed effects model, which controls for all of those criteria (which are by their nature time invariant over the course of the survey). Thus, sample weights are not employed in the main regression analyses. Because there are still differing v iewpoints on using weights in an analysis such as this, I present a weighted regression analysis in the appendix 15 Calculating Rate of R eturn The preceding model s will estimate the change in wages due to a near completer returning to finish his or her baccalaureate degree. Under a strict set of assumptions, such model s rate of return: 1) there are no costs for schooling (direct or indirect); 2) schooling does not affect length of work life; 3) there are no taxes on additional in come earned; and 4) additional schooling does not affect the impact of experience (or other variables) on earnings ( Bjorklund & Kjellstrom, 2000; Heckman, Lochner, & Todd, 2008). Because these assumptions do not hold up in reality (particularly for near co mpleters), calculating t he actual return on investment r equires additional calculations 15 As is discussed in detail in the appendix, the weighted regression shows a slightly smaller effect size for degree completion, though it maintains its strong statistical significance. It is not possible to test whether there is a statistically significant difference between the two effect sizes.

PAGE 95

84 M pursuing additional education later in life, when they may have fewer years in which to recoup costs. Students who pursue higher education directly from high school have a long working life which allows them to earn back the investment (both in direct and indirect costs) for college attendance. For those deciding whether to enroll later in li fe, it could be that shorter potential earning windows make it difficult to realize a positive return on their investment. Approximations of a return to degree completion should project out over time to account for varying time remaining in the workforce. For the purposes of this study, taxes are ignored as research suggests the progressive tax rate in the United States can reduce the income premium for degree completion, but its overall impact is limited ( Heckman, Lochner, & Todd, 2008 ). The progressive t ax rates that are a feature of t he United States tax code mean that the higher wages earned by near completers who finish degrees are taxed at higher rates, reducing its overall benefit. However, Heckman, Lochner, & Todd (2008) find that accounting for the tax code reduces wage premiums by less than one percentag e point, so for the purposes of this study, taxes are ignored. I begin the analysis of whether near completers who finish a degree receive a positive economic return with the broader model original ly proposed by Becker (1993) and adapted from Heckman, Lochner, & Todd (2008): where Y 1 represents the earnings for an individual with a college degree at time t while Y 0 represents the earnings for a near completer at time t. r represents a discount rate,

PAGE 96

85 while represents the number of years an individual takes to complete a degree The first term in the numerator sums discounted earnings from t= to T 1 which is the retirement age for an individual with a college degree. The second term in the numerator and the denominator sums discounted earnings from t=0 to T 0 which is the retirement age for a near completer. The direct costs for attending college are included in the term C t where C represents costs for education in year t, so if the individual is not enrolled in that year, this term will be zero. As noted ear lier, this model only accounts for foregone wages by assuming that the individual does not work at all for years, by beginning the summation in the year rather than year 0. I modify this model by first assuming that individuals will work until the sam e age Second, I explicitly include foregone wages as a separate term (F t ) The modified equation is as follows: The equation can be further simplified by substituting the wage pre mium, written as a percentage of Y 0 in place of the Y 1 ,t term and carrying out a similar procedure for the F t term. This information can be supplied through the estimates of wage premiums and foregone wages from the fixed effects model. The remaining unknown variables will be the direct cost of attendance and the length of time remaining in the workforce. Because there are not data on the actual tuition costs paid by individual near s. Instead, I present a range of scenarios where tuition is calculated as a percentage of an average near This helps simplify the above equations so that the coefficient for

PAGE 97

86 wage premiums, the direct costs, and the coefficient for for egone wages are all presented as a percentage of earnings. Given th e numerous potential avenues for discounted tuition (including grants, scholarships, employer assistance, and military benefits as well as part time enrollment ) I use a range of different tuition figures to calculate these potential average reported tuition at different points in time and calculate that tuition figure as a percentage of average earnings Rather than producing a definitive rate of return for individuals attending different types of schools, this approach will produce an estimated rate of return dependent on assumptions about the level of direct costs borne by an individual. While this a pproach is imperfect in some ways, it is superior to ignoring direct costs altogether. The final unknown variable s are the length of time remaining in the workforce and the number of years an individual is enrolled A near completer who finishes a degree and only works for one year following graduation would likely not see a positive economic return (unless the wage premium was extremely large). It may take many years of employment for the wage premium to cover the costs borne by the individual, particula rly when using an appropriate discount rate. Additionally, a near completer who takes several years to graduate and bears substantial foregone wages during these years, will also likely earn a smaller economic return, although a part time student may foreg o fewer wages. These different contingencies are accounted for in the dif ferent scenarios presented in the results chapter below Plotti ng the results produces a curve with a negative initial return (when a near completer is paying direct costs and sacrificing wages) that increases over time Whether

PAGE 98

87 the plot becomes positive, and if so, how long it takes and how high it gets will provide clarity on the economic return for near completers who finish degrees. Although such plots are not common in other literature on the returns to education, they make sense for the population under consideration here. Similar plots are used in investing when there may be negative returns early in the investment period, followed by a period of increasing returns. 16 A hypothetical plot is included below. In this example, year 0 represents the decision to return to finish a degree. This individual remains enrolled for two years, at which point he or she has received a substantially negative return of about 25 percent. F ollowing an assumed graduation at year two, the return begins to climb as the individual earns a wage premium for his or her degree, crossing the break even threshold at about year 5. The return levels off at just over 10 percent. In the results section, I present a series of similar plots approximating the return to degree completion under a series of different tuition scenarios. 16 Curve: A Primer on Interim Performance of Private Equity http://www.goldmansachs.com/gsam/pdfs/USTPD/education/understanding_J_Curve.pdf Figure 7 : Plot of hypothetical rate of return for near completer finishing a degree

PAGE 99

88 Event History A nalysis characteristics affect the probability that he or she will return to finish a degree, I employ an event history model. A standard probability analysis using a probit or logistic regression mo del is not appropriate for the s e longitudinal data because of two varying explanatory variables (Allison, 1982; Mills, 2011). Censorship refers to the fact that even though a near completer did not graduate during the survey period, there is still a possibility that he or she might graduate during subsequent periods (Allison, 1982; Mills, 2011). Further, in a probit or logistic regression model, an individual who becomes a near completer late in the survey period but never finishes a degree would have as much impact on the estimations as someone who became a near completer early in the survey period and remained so for 25 years. Additionally, these approaches cannot account for changes in explanatory variables over time (Allison, 1982; Mills, 2011). T hese shortcomings are all relevant for the anal yses in quest ion. Essentially, the event history analysis employed here calculate s how different variables affect the likelihood that the event of interest will occur (in this case, the event is the decision to return to school for an enrollment spell that results in degree completion.) 17 Owing to its origins in medical and biological research, this methodology is also known as s urvival analysis. T he occurrence of the event of interest is known as a (which in the case of medical research often means the death of a subject) and 17 Due to the fact that many variables of interest particularly income are affected by enrollment status, it would not be appropriate here to model graduation as the event of interest.

PAGE 100

89 t he time until that event occurs is known as the survival time. 18 In this case, a longer surv ival time means that an individual who has become a near completer is taking longer to return to finish his or her degree. There are multiple opti ons for selecting a model to estimate the impact of different variables on the likelihood of graduation, including semi parametric models such as the Cox proportional hazards model and parametric models The latter set of models can be carried out in two d ifferent forms: accelerated failure time (AFT), which estimate s the effect of covariates on increasing or decreasing survival times, and pr oportional hazards models which estimate how different cov rate. This is the likelihood o f failure occurring in a time period given that it has not occurred prior to that (Bradburn, Clark, Love, & Altman, 2003; Jenkins 2008; Mills, 2011 ; Taniguichi & Kaufman, 2005 ). Additionally, within the family of parametric models, there are numerous choices for specifying the distribution of the hazard over time, including exponential, Weibull, Gompertz, log logistic, log normal, and gamma models (Bradburn, Clark, Love, & Altm an, 2003; Jenkins 2008; Mills, 2011). Choosing between these approaches depends on theoretical insights into the shape of the hazard rate, supported by a series of post estimation diagnostics (Box Steffensmeier & Jones, 2004; Bradburn, et al., 2003; Jenkin s, 2008). The Cox proportional hazards model does not require this specification between distributions and is consequently a popular approach in the literature (Bradburn et al., 2003 ; Mills, 2011; Orbe, Ferreira, & Nunez Anton, 2001). Although I employed this model initially, as discussed in greater detail below, post estimation diagnostics showed 18 ortunate, I use it here because it is relevant to the choice of an Accelerated Failure Time (AFT) model.

PAGE 101

90 that some of the covariates violate the proportional hazards assumption, which is a necessary criterion for employing a Cox model. Instead, I selected from amon g the parametric models, and began with the assumption that the hazard rate for near completers to finish degrees initially increases and reaches a peak before decreasing the remaining years Based on this assumption, a log normal model is appropriate (Bra dburn, et al., 2003; Jenkins, 2008; Mills, 2011). P ost estimation diagnostics support this choice. The model uses an AFT approach, so survival times are the dependent variable (Mills, 2011). Covariates that are associated with shorter survival times can th us be said to be associated with an increase in the likelihood that a near completer will return to finish his or her degree. The model, from Mills (2011) and others, is specified as follows: l n(t j ) = 1 x i1 + 2 x i2 + where the dependent variable is the natural logarithm of the survival time ( t j ) ; x i1 is a vector of individual level time invariant variab les, including race/ethnicity, gender, access to a library at the outset of the survey, access to newspapers and magazines at the outset of the survey parental education, family socioeconomic status (as measured by poverty stat us at the outset of the survey), and score on the ASVAB as a rough measure of ability ; and x i2 is a vector of individual level variables that vary over time including income, nu mber of children, health status age, military status (as a potential control for GI Bill benefits that could increase the likelihood of returning), and experience The se covariates are generally shown throughout the literature to affect levels of educational attainment. Further, the descriptive data comparing near completers who never finished and those that did also show s that there may be statistically significant differences for m any of these.

PAGE 102

91 All of the time variant variables are lagged to ensure that causal variables are measured prior to the outcome of interest. Failure to do so could introduce significant reverse causation problems to the results (Box Steffensmeier & Jones, 20 04; Mills, 2011). With the event of interest being the beginning of the enrollment spell that leads to graduation it would not be appropriate to use these time variant variables measured in the same year an individual reports being enrolled to explain cha nges in the likelihood of enrollment Income, for example, might be lower while the individual is enrolled, but increase significantly following the completion of a degree. A n analysis of the impact of income in the year of enrollment would likely show tha t lower incomes are associated with a higher likelihood of enrolling when in fact the income change is due to foregone wages. Instead, I lag these variables to the survey round prior to the beginning of the enrollment spell that leads to degree completio n. Methodology: Conclusion Each of these approaches tests hypotheses associated with different research questions. The hypotheses associated with each methodological approach are detailed in Table 15 The use of individual fixed effects alleviates concerns about omitted variable bias for time invariant individual characteristics, but there are still some potential limitations associated with this methodological approach. There could still be difficult t o measure characteristics (such as maturity) that could change over time and be correlated with both earnings and the likelihood of returning to school. An individual could have left postsecondary education near a degree and started working, but years late r, developed stronger motivation and ambition towards work that ultimat ely results in higher wages as well as le a d ing him or her to return to school to complete a degree. There could also be

PAGE 103

92 Table 15 : Methodological approaches used to test hypotheses Analytic Method Hypothesis Individual Level Fixed Effects Regression H1.1: Near completers who return to postsecondary education to finish a baccalaureate degree will earn a wage premium compared to those who do not return The premium is sufficiently large to result in a positive economic return. l depend on years remaining in the workforce and costs of attendance. H1.2: The benefits of degree completion by near completers do not accrue equally to all population sub groups. H2: Individuals who finish a degree at a public college or university will realize a higher economic return than those w ho finish a degree at a non profit or for profit college or university. Event History Analysis H3: Factors that are important in pre dicting educational successes for traditional students will also be important in predicting whether near completers finish degrees. observable life events that may be strongly related to both income and desire to obtain more schooling, such job changes to fields that pay more and offer opportunity for advancement based on education It may be possible to identify observable characteristics that are related to maturity, but this would likely be highly speculative in nature. Even with these limitations, t he individual level fixed effects model provides strong controls for fa ctors that bias many estimates of the link between education and earnings. This provides a much stronger causal link between changes in education status and an U sing event history analysis is a substantial improvement from trying to estimate which factors are associated with degree completion using a probit or logistic regression model. This approach will not only better fit the data employed, but also allow for t he estimation of characteristics that can change over the course of the survey. While it does

PAGE 104

93 provide a slightly more complicated interpretation, this approach better capture s the impact of certain characteristics on the likelihood that an individual retur ns to finish a degree.

PAGE 105

94 CHAPTER V RESULTS The models described above produce results showing that some, but not all, near completers are likely to earn a positive economic return for baccalaureate degree completion. Further, many of the individual level factors that are traditionally important predictors of education outcomes do not appear to affect whether near completers return to finish degrees This section presents the results for all forms of the models employed. Discussion of the results is include d in the subsequent chapter. The Wage Premium for Degree Completion The results for the individual level fixed effects regression show how different factors including education attainment status first three models es timate the effect of degree completion regardless of the sector of the school from which the individual graduates. F ive models are presented in Table s 16 20 Model (1) shows the coefficients for the entire sample while model (2) show s the results when sel ect interaction term s for gender are included model (3) shows the results for different racial/ethnic groups model (4) shows the results by initial SES status, and model (4) show s results for public, private non profit, and private for profit graduates. Post regression diagnostics. I conduct several diagnostic tests to ensure that the models are appropriate First, I conduct a Hausman test to determine whether the fixed effects approach used here is appropriate, or whether a random effects model may be used (Allison, 2009; Hausman, 1978) The test confirms with high levels of significance (p<.00 0 1), that I should use a fixed effects mo del. With a fixed effects model, there are two tests of joint significance. The test for the joint significance of the included

PAGE 106

95 covariates shows that I can reject at the highest levels of significance (p<.0001) that the coefficients are jointly equal to ze ro. Fixed effects models also provide an F statistic that tests whether the effect of the time invariant individual characteristics are equal to zero. I can also reject at the highest levels of significance (p<.0001) the null hypothesis that the time invar iant characteristics jointly have no effect on the dependent variable The fixed effects regression also necessitates testing for heteroskedasticity, which, if present, would violate the assumption that the variance of the error is equally distributed acro ss individuals ( Angrist & Pischke, 2009; Baum, 2001 ; Wooldridge, 2006 ) Testing this assumption with a modified Wald test however shows that I must reject this hypothesis and conclude that heteroskedasticity is present (Baum, 2001) I also examine the ex tent to which the error terms are correlated across time, a phenomenon known as serial correlation, using a test developed by Wooldridge (2002). The test shows I must reject the null hypothesis that there is no serial correlation present. To correct for th e presence of both heteroskedasticity and serial correlation, I use clustered robust standard errors (Baum, 2001; Drukker, 2003). Another key assumption for this regress ion is that there are not linear relationships b etween the covariates a phenomenon k nown as m ulticollinearity I examine the collinearity of the covariates and identified two variables that were highly correlated. I initially included one control variable for the number of weeks that an k he or she could do and one for the number of weeks health status limited the type of work he or she could do. Both were included to control for the fact that health status might be correlated with both finishing a degree and wage levels. Based on the very high level of correlation between the two

PAGE 107

96 variables (r>.95) I include only the variable for limitations on the amount of work (which is not statistically significant). 19 Income benefits for degree completion Th e model s that follow use near completers who are neither enrolled nor enrolled in the following survey round as the comparison group. The key comparisons are with the other three education levels: pre enrollment, which is included to test and control for t he presence of the Ashenfelter Dip; enrollment, which includes anyone enrolled in college; and near completers who have graduated. Model (1) show that pre enrollment near completers on average have wages that are seven percent higher tha n the comparison gr oup. 20 This coefficient is statistically significant at p<.05. Those who are enrolled have 16 percent lower wages than the comparison group, while finishing a degree leads to a 1 7 percent wage increase. 21 T hese income differences are statistically significan t at p<.0 1. Age and experience both show statistically significant coefficients in line with the literature, with wages increasing as individuals age and gain more experience, but the rate of increase slowing down over time. The coefficients for unemployme nt and being out of the labor force are also both statistically significant, with each additional week of either leading to three percent lower wages. These results are statistically significant at p<.01. Health status, number of children, and marital stat us are not statistically significant for the full sample. 19 I also tested including the health variable for number of weeks health status limited the type of work. It was not significant either and did not not iceably affect other coefficients. 20 The wage coefficients from the fixed effects models represent the effect of a unit change in the variable on the natural logarithm of wages. To convert the coefficients to the percentages used here, the following operat ion is used: Pct. chg = ( e coefficient 1) x 100 21 This wage premium represents the average effect over time for degree completion. An estimation that examines how the premium changes over time is included in the appendix. That analysis shows that the eff ect of degree completion appears to start lower than this average effect and increase over time.

PAGE 108

97 Table 16 : Effects of education status on earnings Variable (1) All Obs Coefficient Std. Err Pre enrollment a 0.07** 0.03 Enrolled a 0.17*** 0.03 Graduate a 0.16*** 0.05 Age 0.12*** 0.03 Age 2 0.001*** 0.0004 Exp 0.29*** 0.04 Exp 2 0.03*** 0.01 Exp 3 0.001*** 0.0003 Exp 4 0.00002*** 0.000007 Wks unemployed 0.03*** 0.002 Wks out of labor force 0.03*** 0.002 Health limitations 0.02 0.02 Number of children 0.02 0.02 Never married b 0.08 0.11 Married b 0.14 0.11 Separated b 0.12 0.12 Divorced b 0.13 0.11 Observations 12,156 1,042 Individuals Note: Year fixed effects also included but not reported. Significant at p<.1; ** Significant at p<.05; *** Significant at p<.01 a Reference category is near completers who are not enrolled, have not finished a degree, and do not enroll the following year. b Reference category is widowed Differential effects by gender. M odel 2, the results of which are shown in the T able 17 presents the differential impacts of gender 22 There is not a statistically significant difference in the effects of education level by gender. There is a statistically significant difference in the effect that children have on earnings. F or women, the 22 Note that in model 2, as well as subsequent fixed effects models that use interaction terms, I do not include a categorical variable for the time invariant c haracteristic that is part of the interaction term (in this case gender). Because the time invariant characteristics are absorbed by the fixed effects model, it is not feasible to include these variables separately. This approach follows Allison (2009). Th e coefficients for the interaction terms can be interpreted in the same way they would be in a model without fixed effects (Allison, 2009).

PAGE 109

98 coefficient is .13 log points lower than the effect for men, suggesting that the coefficients point in the opposite direction. The only other statistically significant differences are that for women, each additional week out of the labor forc e has a slightly smaller ( .01 log points ) effect on earnings than for men, while each additional week of unemployment has a slightly larger ( .01 log points ) effect. Differential effects by racial/ethnic background. Model 3, wi th results presented in Table 18 shows the differential effects of degree completion by racial/ethnic background. The racial/ethnic categories for Hispanics and African Americans are combined into one category and compared to the category for other ethnic ities (which is likely predominantly W hite individuals). The interaction terms include the variable for other ethnicities, so they show the difference in the effect on each variable for those individuals compared to African Americans/Hispanics. 23 T he result s show statistically significant differences for some categories of education status. African American/Hispanic near completers receive a highly significant (p<.01) 31 percent income premium for degree completion T he coefficient for degree completers from other racial/ethnic backgrounds is .20 log points lower suggesting that the effect for this group is not significantly different from zero The difference between the two coefficients is highly statistically significant (p<.01). The coefficient for enrol lment suggests that African Americans/Hispanics forego about 11 percent of their wages when enrolled (p<.05). The coefficient for other ethnicities is .09 log points lower, suggesting that they 23 The conventional approach would be to use the other ethnicities category as the base terms and the African America n/Hispanic group in the interaction variables. However, the results, in which African Americans/Hispanics receive a statistically significant wage premium for graduation and the other ethnicities group does not (though it is statistically different relativ e to African Americans/Hispanics), are easier to interpret using this specification.

PAGE 110

99 forego about 18 percent of their wages. This difference shows modest statistical significanc e (p<.10). There is no statistically significant difference in the wage increase in the year prior to enrollment between the two groups. The only other interaction term with statistical significance shows that the effect of ea ch week out of the labor force is associated with one percentage point lower earnings for other ethnicities compared to African Americans/Hispanics. Differential effects by initial poverty status. Model 4 shows differential effects of ilial poverty status in 1979 on earnings. As can be seen in Table 19, none of the interaction terms show statistical significance. The implications of this are discussed in detail in the following chapter. Public vs. non profit vs. for profit graduates. To estimate the differences in the return to degree completion by sector, I have constructed a series of education status variables that incorporate the type of school an individual graduated from with the education status variables used in previous models Interaction terms are included for each status level for those who eventually graduate from private, non profit and for profit schools. The results show that there are not statistically significant differences in the wage premium by sector for graduates, nor are there statistically significant difference s among the other education status variables by sector.

PAGE 111

100 Table 17 : Differential wage premiums by gender Variable ( 2 ) Gender Interactions Coefficient Std. Err Pre enrollment a 0.09* 0.05 Enrolled a 0.2 0 *** 0.04 Graduate a 0.17*** 0.06 Pre enrollment x female a 0.06 0.07 Enrolled x female a 0.06 0.06 Graduate x female a 0.06 0.09 Age 0.14*** 0.04 Age 2 0.002*** 0.0004 Age x female 0.05 0.04 Age 2 x female 0.001 0.0005 Exp 0.22*** 0.05 Exp 2 0.02*** 0.01 Exp 3 0.001*** 0.0004 Exp 4 0.00002** 0.000010 Exp x female 0.12 0.07 Exp 2 x female 0.01 0.01 Exp 3 x female 0.001 0.001 Exp 4 x female 0.00001 0.00001 Wks unemployed 0.03*** 0.002 Wks out of labor force 0.02*** 0.003 Wks unemployed x female 0.01* 0.003 Wks out of labor force x female 0.01*** 0.003 Health limitations 0.06 0.04 Health limitations x female 0.06 0.05 Number of children 0.05 ** 0.02 Number of children x female 0.13*** 0.03 Never married b 0.07 0.34 Married b 0.28 0.33 Separated b 0.34 0.34 Divorced b 0.15 0.35 Never married x female 0.07 0.36 Married x female 0.22 0.35 Separated x female 0.32 0.36 Divorced x female 0.01 0.37 Female observations 563 Male observations 479 Notes: Year fixed effects also included but not reported. Significant at p<.1; ** Significant at p<.05; *** Significant at p<.01 a Reference category is near completers who are not enrolled, have not finished a degree, and do not enroll the following year. b Reference category is widowed.

PAGE 112

101 Table 18 : Differential effects by racial/ethnic background Variable (3) Racial/Ethnic Interactions Coefficient Std. Err Pre enrollment a 0.09* 0.05 Enrolled a 0.12** 0.04 Graduate a 0.27*** 0.06 Pre enrollment x other ethnicity a 0.04 0.07 Enrolled x other ethnicity a 0.09* 0.06 Graduate x other ethnicity a 0.2 0 ** 0.09 Age 0.11*** 0.04 Age 2 0.002*** 0.0004 Age x other ethnicity 0.01 0.05 Age 2 x other ethnicity 0.0002 0.0005 Exp 0.31*** 0.05 Exp 2 0.03*** 0.01 Exp 3 0.002*** 0.0004 Exp 4 0.00003*** 0.00001 Exp x other ethnicity 0.04 0.07 Exp 2 x other ethnicity 0.01 0.01 Exp 3 x other ethnicity 0.0004 0.001 Exp 4 x other ethnicity 0.00001 0.00001 Wks unemployed 0.03*** 0.002 Wks out of labor force 0.02*** 0.003 Wks unemployed x other ethnicity 0.003 0.003 Wks out of labor force x other ethnicity 0.01** 0.004 Health limitations 0.02 0.03 Health limitations x other ethnicity 0.01 0.04 Number of children 0.003 0.02 Number of children x other ethnicity 0.04 0.03 Never married b 0.05 0.16 Married b 0.12 0.15 Separated b 0.04 0.16 Divorced b 0.1 0.16 Never married x other ethnicity 0.08 0.22 Married x other ethnicity 0.07 0.21 Separated x other ethnicity 0.26 0.24 Divorced x other ethnicity 0.08 0.23 African Americans/Hispanics: 477 Other Ethnicities: 565 Notes: Year fixed effects also included but not reported. Significant at p<.1; ** Significant at p<.05; *** Significant at p<.01 a Reference category is near completers with no degree who are not enrolled nor enroll the following year. b Reference category is widowed.

PAGE 113

102 Table 19 : Differential effects by 1979 poverty status Variable (4 ) Poverty status interactions Coefficient Std. Err Pre enrollment a 0.09** 0.04 Enrolled a 0.15*** 0.03 Graduate a 0.14*** 0.05 Pre enrollment x 1979 poverty status a 0.12 0.1 0 Enrolled x 1979 poverty status a 0.02 0.07 Graduate x 1979 poverty status a 0.15 0.13 Age 0.12*** 0.03 Age 2 0.001*** 0.0004 Age x 1979 poverty status 0.04 0.05 Age 2 x 1979 poverty status 0.001 0.001 Exp 0.27*** 0.04 Exp 2 0.03*** 0.01 Exp 3 0.001*** 0.0004 Exp 4 0.00003*** 0.00001 Exp x 1979 poverty status 0.05 0.08 Exp 2 x 1979 poverty status 0.01 0.01 Exp 3 x 1979 poverty status 0.0004 0.001 Exp 4 x 1979 poverty status 0.00001 0.00002 Wks unemployed 0.03*** 0.002 Wks out of labor force 0.03*** 0.002 Wks unemployed x 1979 poverty status 0.004 0.004 Wks out of labor force x 1979 poverty status 0.002 0.004 Health limitations 0.03 0.03 Health limitations x 1979 poverty status 0.04 0.05 Number of children 0.03 0.02 Number of children x 1979 poverty status 0.04 0.04 Never married b 0.11 0.11 Married b 0.19* 0.11 Separated b 0.17 0.12 Divorced b 0.22* 0.11 Never married x 1979 poverty status 0.09 0.33 Married x 1979 poverty status 0.03 0.33 Separated x 1979 poverty status 0.03 0.36 Divorced x 1979 poverty status 0.09 0.32 Individuals not in poverty, 1979 780 Individuals in poverty, 1979 208 Notes: Year fixed effects also included but not reported. Significant at p<.1; ** Significant at p<.05; *** Significant at p<.01 a Reference category is near completers with no degree who are not enrolled nor enroll the following year. b Reference category is widowed.

PAGE 114

103 Table 20 : Income differences by sector of graduation Variable (5) Sector effects Coefficient Std. Err Pre enrollment a 0.08 0.08 Enrolled a 0.31 *** 0.1 0 Graduate a 0.13 0.09 Yr. prior to enr. x non profit a 0.08 0.17 Yr. prior to enr. x for profit a 0.29 0.21 Enrollee x non profit a 0.07 0.19 Enrollee x for profit a 0.06 0.24 Graduate x non profit a 0.18 0.18 Graduate x for profit a 0.22 0.17 Age b 0.12*** 0.03 Age 2 b 0.001*** 0.0004 Exp b 0.26*** 0.04 Exp 2 b 0.02*** 0.01 Exp 3 b 0.001*** 0.0004 Exp 4 b 0.00002*** 0.00001 Wks unemployed 0.03*** 0.002 Wks out of labor force 0.03*** 0.002 Wks unemployed x non profit 0.01 0.01 Wks unemployed x for profit 0.06 0.03 Wks out of labor force x non profit 0.004 0.01 Wks out of labor force x for profit 0.02*** 0 .004 Health limitations 0.01 0.03 Health limitations x non profit 0.03 0.07 Health limitations x for profit 0.16 0.07 Number of children 0.02 0.02 Number of children x non profit 0.07 0.07 Number of children x for profit 0.02 0.07 Never married b c 0.07 0.11 Married b c 0.13* 0.1 Separated b c 0.11 0.11 Divorced b c 0.12 0.11 Public Graduates Obs.=2,430 N=197 Non profit Graduates Obs.=827 N=74 For Profit Graduates Obs.=92 N=13 Notes: Year fixed effects also included but not reported. The interaction terms combine the dichotomous variable for sector of graduation with enrollment/graduation status. Significant at p<.1; ** Significant at p<.05; *** Significant at p<.01 a Reference category is near completers who are not enrolled, have not finished a degree, and do not enroll the following year. b Interaction terms for age, experience, and marriage eliminated due to multicollinearity. c Reference category is widowed.

PAGE 115

104 Returns to Degree Completion The results from the preceding section show generally that near completers who finish a degree earn a wage premium, though the model with race/ethnicity interaction terms calls into question whether all population sub groups a re likely to earn that premium. Additionally, the models generally show that near completers who return to school bear some indirect costs in the form of foregone wages. In addition to these two factors, accurately estimating the return on degree completio n requires some knowledge about the direct costs borne by those who return to finish degrees. Without data on the actual tuition paid (which do not exist in the NLSY) it is not possible to directly calculate he sample. Instead, I use the average annual tuition prices published by the National Center for Education Statistics as a starting point for approximating direct costs. Given the potential for returning adults to receive financial aid in the form of gove rnment grants, employer assistance, institutional tuition discounting, and other benefits such as those offered as part of the GI Bill, as well as the likelihood that many adults enroll part time rather than full time, I calculate several scenarios using d ifferent fractions of the average tuitions. The resulting approximations show the internal rate of return over time given the assumption around tuition prices. Direct costs. To approximate direct costs, I reprise the average tuition data and calculate it as a percentage of mean earnings for near completers The results of this are depicted graphically in the figure below. To develop the scenarios for rates of return, which are dependent on tuition as a percentage of earnings representing direct costs, I use percentages at different time points. Although wages tend to climb over time, due to

PAGE 116

105 experience and other factors, tuition grows unevenly, at times outpacing natural wage growth. The purpose of these tuition approximations is not to try to exactly capture the direct costs for each individual, but to provide a reasonable basis for the assumptions I make. Actual tuition paid is also dependent on wide variations in the availability of financial aid, employer tuition assistance, and other benefits that can reduce tuition costs, as well as differential tuitions by the school to which a near completer returns. Additionally, as will b e seen below, I use a range of estimates, based on quarter half and full price for each sector to approximate the likely part time enrollment of near completers. Note: All calculations completed using adjusted 2012 dollars and appropriately weighte d income averages

PAGE 117

106 Figure 8 : Public tuition as a percentage of mean wages of near completers Note: All calculations completed using adjusted 2012 dollars and appropriately weighted income averages Figure 9 : Non profit tuition as a percentage of mean wages of near completers Note: All calculations completed using adjusted 2012 dollars and appropriately weighted income averages For profit tuition not calculated separately until 2000. Figure 10 : For profit tuition as a percentage of mean wages of near completers Because I also intend to approximate returns for African Americans/Hispanics, who showed a large and significant wage premium in the interaction model, I repeat the calculations limi ted to that racial/ethnic group for public tuition 24 Due to slightly lower 24 Due to small subsample sizes (i.e. African American/Hispanics who graduated from for profit schools are in the single digits), calculations are not performed for subgrou ps by sector. Instead, average public

PAGE 118

107 average incomes, tuition for the African American/Hispanic group would account for a larger percentage of earnings. Note: All calculations completed using adjusted 2012 dollars and appropriately weighted income averages Figure 11 : Public tuition as a percentage of earnings of African Americans/Hispanics Approximations of economic returns to degree completion. Even with the models for wage premiums above, a series of assumptions are necessary to calculate the economic return. I begin by using a discount rate of five percent. Within social sciences literature, there is a range of discount rates empl oyed. Mincer (1958) in the foundational work on returns to education, uses a rate of four percent Cellini (2012) uses three percent in a study on the costs and benefits of for profit education; Hoxby and Turner (2013) use a similar rate in a study on low income students and education gains. Other research on education uses average bond prices choose over the period in question to arrive at a discount rate of 7.4 percent (Banzhaff and Bhallah, 2012). Thus, while there is not a tuition is used to approximate returns for this sub group because the majority of near completers who graduate do so from public institutions.

PAGE 119

108 strong consensus on the appro priate discount rate to apply, the choice of five percent is within the range of other estimates, and alternative discount rates have modest effects. 25 I also make an assumption about the length of time required to complete a degree and use the medi an length of a near completer s final enroll ment spell, which is two years. This implies an assumption that near completers make one final return to school after attaining near completer status. 26 I do not make assumptions about the length of time an individu al has remaining in the workforce, but the plots of economic returns presented below carry the estimations out to 40 years post enrollment. Using these assumptions, I produce a series of approximations of the return to baccalaureate degree completion by ne ar completers over time. The series consists of scenarios based on tuition by sector at two different points of time for the general model. I also produce approximations for African Americans/Hispanics at one point in time. The percentages for the rate of return represent an annual return over (or under) what a near completer who did not return to finish a degree would earn. The results of each scenario are depicted in the figures below, listing the year used for tuition calculations and the values for for egone wages and wage premium s for degree completion. As seen below, the rates of return vary somewhat depending on the length of time remaining in the workforce. The figures show the rates of return over t ime, with the initial year being the first year of final enrollment. As would be expected the rate of return is 25 Comparing results from the approximations when using three and seven perc ent discount rates shows that the eventual return to degree completion would change by about 1.2 percentage points (higher with the three percent rate and lower with the seven percent rate) after 40 years. Given that the rates of return accumulate over a l ifetime, this is a modestly significant difference. 26 True enrollment patterns of near completers are more complex and the pattern of multiple reenrollments and stopouts warrants further study. Further, calculating returns assuming longer enrollment spells has predictable effects lengthening the years until the return turns positive and decreasing the longer term average annual return.

PAGE 120

109 negative during the years in which the individual is enrolled, before climbing post graduation. The year in which the return turns positive intuitively changes depending on the direct costs. The returns are calculated using the formula adapted from Becker (1993) and Heckman, Lochner, and Todd (2008). The return is calculated by summing the annual wages of a near completer (less direct and indirect costs) plus the post graduat ion wage premium (appropriately discounted) divided by the annual wages of a near completer who has not finished a degree plus costs. In the first scenario, I approximate the returns to education for a near completer returning to a public college for two y ears using the wage premium and foregone wages results from m odel 1. 27 For direct costs, I assume full time tuition is nine percent of earnings (calculated from the average tuition rates for 1990 as a percentage of average near completer wages in the same y ear). The results represent the annualized internal rate of return any number of years after the decision to reenroll. So in the figure below, two years after returning, a full time student would have realized an average annual return of negative 20 percen t. Ten years after reenrollment, the individual would realize an average annual re turn of nearly seven percent. 27 Two years is the median length of enrollment for near completers who finish baccalaureate degrees during their final spel l of schooling. I include additional scenarios in the appendix lengthening the enrollment period and decreasing the indirect costs under the assumption that there are likely some individuals who attend school on a limited basis and may not forego the same level of wages. Such scenarios are discussed in greater detail in the appendix.

PAGE 121

110 Wage premium: 17% (from Model 1 results) Indirect costs: 16% (from Model 1 results) Tuition rates: 1990 Figure 12 : Return for baccalaureate degree completion public tuition rates These results show, given the numerous assumptions made above, that an average near completer who finishes a baccalaureate degree paying full public tuition for two years, would see a positive economic return by year six and a 10.6 percent return 20 years after reenrollment. An individual paying a quarter of average tuition, would see a positive economic return in year five, and would see a return of about 11.7 percent 20 years after re enrollment. Figure 13 shows similar calculations for non profit/for profit average tuition rates. 28 These results show that individuals paying full price average tuition would not earn a positive return until 11 years after reenrollment. Twenty years after reenrollment, the average individual would see a 5.3 percent positive return. Individuals paying a quarter of full price would see a positive return by year six and a 10.3 percent gain by year 20. 28 As noted above, non profit and for profit tuitions were not calculated separately until 2000.

PAGE 122

111 Wage premium: 17% (from Model 1 results) Indirect cos ts: 16% (from Model 1 results) Tuition rates: 1990 Figure 13 : Return for baccalaureate degree completion non profit tuition rates Examining the returns using assumptions for direct costs based on more recent tuition data shows l argely similar results. Although tuition costs generally increased slightly as a 2010, the growth does not substantively affect the approximations. Figures 14 16 show the approximated returns based on public, non profit, and for profit tuition rates, again using quarter half and full time tuition for direct costs. The results show that individuals paying full average public school tuition prices would again see a positive return six years after reenrollme nt and a 9.9 percent average annual return 20 years after reenrollment. An individual paying the equivalent of quarter price tuition would see a positive return after five years and an 11.5 percent return 20 years after reenrollment. Using the average non profit tuition prices (Figure 15) an individual paying full tuition would not see a positive return until 13 years after reenrollment and would only earn a 4. 3 percent return after 20 years. At one fourth of full

PAGE 123

112 non profit tuition, the breakeven point wo uld come at year six and an individual would see a 10.0 percent return after 20 years. Using average full for profit tuition rates (Figure 16), the average individual would see a positive return by year 9 and a 7.7 percent average annual return by year 20, compared to year 5 and 11.0 percent for one quarter tuition rates. For ease of comparison, the breakeven point and 20 year return for all approximations in this section are collected in Table 21 at the end of this section. Wage premium: 17% (from Mod el 1 results) Indirect costs: 16% (from Model 1 results) Tuition rates: 2010 Figure 14 : Return for baccalaureate degree completion public tuition rates

PAGE 124

113 Wage premium: 17% (from Model 1 results) Indirect costs: 16% (from Model 1 results) Tuition rates: 2010 Figure 15 : Return for baccalaureate degree completion non profit tuition rates Wage premium: 17% (from Model 1 results) Indirect costs: 16% (from Model 1 results) Tuition rates: 2010 Figure 16 : Return for baccalaureate degree completion for profit tuition rates

PAGE 125

114 Rates of return for African Americans/Hispanics. The results of the interaction model for wage premiums by racial/ethnic group (Model 3) show that A frican Americans/Hispanics earn a higher wage premium than is evident in the model t hat considers the whole sample. Repeating the process used to derive the general approximations on returns to baccalaureate degree shows that African Americans/Hispanics, d espite the fact that tuition takes up a larger percentage of their earnings based on the sample averages, earn s ubstantially higher returns. The higher relative direct costs are offset because this group tends to forego fewer wages on average while enrolle d. Given the general similarity between approximations based on 1990 and 2010 tuition levels, only the latter are shown here. As can be seen in Figure 17, based on the results from Model 3, at full public tuition prices, at these wage premiums and indirec t cost rates, African American and Hispanic individuals would see a positive return in year 4 and a 21. 4 percent average annual return after 20 years. At 25 percent of full tuition, these assumptions would yield a positive return in also in year 4 and a 23 .6 percent average annual return after 20 years. Using average non profit tuition rates (Figure 18), at full price an average African American/Hispanic individual would see a positive return by year eight and a 14.1 average annual return after 20 years. A t one quarter price, this changes to year 4 and 21.6, respectively. At full for profit rates (Figure 19), the breakeven point is six years with an 18.6 percent average annual return after 20 years, compared to four years and 22.8 percent at one quarter of average annual tuition rates.

PAGE 126

115 Wage premium: 31% (from Model 3 results) Indirect costs: 11% (from Model 3 results) Tuition rates: 2010 Figure 17 : African American/Hispanic return for baccalaureate degree completion public tuition rates Wage premium: 31% (from Model 3 results) Indirect costs: 11% (from Model 3 results) Tuition rates: 2010 Figure 18 : African American/Hispanic return for baccalaureate degree completion non profit tuition rates

PAGE 127

116 Wage premium: 31% (from Model 3 results) Indirect costs: 11% (from Model 3 results) Tuition rates: 2010 Figure 19 : African American/Hispanic return for baccalaureate degree completion for profit tuition rates The results f or all of the above approximations are collected into the Table 21 summarizing the varied approximations of rates of return across all of the different scenarios employed

PAGE 128

117 Table 21 : Approximated rates of return for baccalaureate degree completion Subsample Tuition Year Sector Tuition Assumption Breakeven year 29 20 yr avg. annual return Full subsample 1990 Public Full 6 10.6% Half 5 11.3% Quarter 5 11.7% Non/For profit Full 11 5.3% Half 8 8.6% Quarter 6 10.3% 2010 Public Full 6 9.9% Half 5 11.0% Quarter 5 11.5% Non profit Full 13 4.3% Half 8 8.0% Quarter 6 10.0% For profit Full 9 7.7% Half 6 9.9% Quarter 5 11.0% African American/Hispanic 2010 Public Full 4 21.4% Half 4 22.8% Quarter 4 23.6% Non profit Full 8 14.1% Half 6 19.0% Quarter 4 21.6% For profit Full 6 18.6% Half 4 21.3% Quarter 4 22.8% Why Do Near Completers Return? The results of the event history analysis are presented in a series of tables below. The models include e stimations for the entire sample and separate models with gender, racial ethnic and poverty interaction terms. The coefficients in these log normal model s relate to the expected time until the event of interest occurs. In th ese models the event of interest is the beginning of an enrollment spell that results in graduation Thus, neg ative 29 would take an average individual to recoup the direct and indirect costs of degree completion given the estimated wage premium and appropriate discounting.

PAGE 129

118 coefficients imply that a unit increase in that variable reduces the survival time or the time un til a nea r completer enrolls to finish his or her baccalaureate degree. Factors that are associated with decreases in survival time can be thought of as potentially increasing the likelihood that an individual will finish a degree. Factors affecting likelihood of r eturn. In the first model, presented in Table 22, without interaction effects, the only time invariant characteristic s that is statistically significant are paternal education and (p<.01) A single grade increase in paternal ed ucation is associated with an eight percent decrease in survival time. 30 A one with a two percent decrease in survival time. Some time varying characteristics also show statistical significan ce. As noted earlier, these variables are lagged to the year prior to enrollment. A one percent increase in earnings is associated with a .23 percent increase in survival time (p<.05). Number of children, weeks out of the labor force, and unemployment show modest statistical significance (p<.10). Each additional child is associated with 22 percent longer survival times; each additional week out of the labor force is associated with a one percent increase; and each additional week of unemployment is associat ed with a three percent increase. The implications of these results and the implications of the non significant covariates are discussed in detail in the subsequent chapter. Gender interactions. Adding interaction terms to account for differences in the effect of cov ariates by gender produces few significant findings, presented in Table 23 30 The coefficients for the survival analysis models represent the effect of a unit change in the var iable on the natural logarithm of survival time. To convert the coefficients to the percentages used here, the following operation is used: Pct. chg = ( e coefficient 1) x 100 cent change in that variable is associated with a percentage change in survival time equal to the coefficient.

PAGE 130

119 Table 22 : Event history analysis of degree completion Variable Coeff icient Std. Err. Time Invariant Characteristics Maternal Education 0.005 0.03 Paternal Education 0.08*** 0.03 Poverty Status (1979) 0.22 0.22 Library Card (1979) 0.24 0.21 HH Rec'd Newspaper (1979) 0.13 0.23 HH Rec'd Magazines (1979) 0.12 0.19 Fem ale 0.02 0.17 ASVAB percentile 0.02*** 0.004 Other Ethnicity 0.03 0.19 Time Variant Characteristics Former military status 0.18 0.3 0 Age 0.04 0.03 Experience 0.04 0.03 Number of Children 0.2 0 0.11 Health Status 0.03 0.17 Log Income 0.23** 0.09 Out of Labor Force 0.01* 0.01 Unemployed 0.03* 0.01 Subjects 952 Failures 241 Notes: ** Significant at p<.01; **Significant at p<.05, *Significant at p<.10. Time variant variables lagged prior to final enrollment spell below The only interaction term with statistical significance is whether the individual received magazines in his/her household growing up. Both the base term and the interaction only show modest significance (p<.10). The base coefficient suggests that for men, receiving magazines is associated with longer survival tim es by a factor of 1.6, while for women it is associated with a 23 percent decrease. Given the marginal significance, caution is warranted before drawing conclusions here. R acial ethnic interactions. Table 24 presents the results with interaction terms fo r racial/ethnic background which again includes few significant findings. Each term is interacted with a categorical variable for non African American/non Hispanic racial The coefficient for receipt of newspapers (p< .10)

PAGE 131

120 and its interaction term (p<.05) show some statistical significance. The base term for suggests that receipt of newspapers by African Americans/Hispanics is associated with 75 percent longer survival times, while for other ethnicities, receipt is asso ciated with a 49 percent decrease. Interactions with familial poverty status. Results from a model that interacts familial poverty status at the outset of the survey are presented below in Table 25 None of the interaction variables show statistical signi ficance, suggesting that there are not differential effects of any of the variables under consideration by the poverty situation in which an individual grew up.

PAGE 132

121 Table 23 : Event history analysis with gender interaction terms Variable Coefficient Std. Err Time Invariant Characteristics Maternal Education 0.03 0.04 Maternal Education x female 0.04 0.06 Paternal Education 0.07** 0.04 Paternal Education x female 0.02 0.05 Poverty Status (1979) 0.05 0.31 Poverty Status (1979) x female 0.29 0.43 Library Card (1979) 0.11 0.27 Library Card (1979) x female 0.15 0.42 HH Rec'd Newspaper (1979) 0.13 0.34 HH Rec'd Newspaper (1979) x female 0.49 0.46 HH Rec'd Magazines (1979) 0.46* 0.27 HH Rec'd Magazines (1979) x female 0.71* 0.37 Female 1.25 2.16 ASVAB percentile 0.03*** 0.01 ASVAB percentile x female 0.01 0.01 Other Ethnicity 0.07 0.27 Other Ethnicity x female 0.21 0.37 Time Variant Characteristics Former military status 0.1 0 0.33 Former military status x female 0.38 0.65 Age 0.04 0.04 Age x female 0.004 0.05 Experience 0.1 0 ** 0.05 Experience x female 0.09 0.07 Number of Children 0.07 0.17 Number of Children x female 0.2 0 0.21 Health Status 0.19 0.43 Health Status x female 0.28 0.46 Log Income 0.33** 0.13 Log Income x female 0.18 0.18 Out of Labor Force 0.02 0.01 Out of Labor Force x female 0.01 0.01 Unemployed 0.05** 0.02 Unemployed x female 0.04 0.03 Male subjects : 440 Male failures : 118 Female subjects : 512 Female f ailures: 123 Notes: ** Significant at p<.01; **Significant at p<.05, *Significant at p<.10. Time variant characteristics lagged prior to final pre enrollment spell.

PAGE 133

122 Table 24 : Event history analysis with racial/ethnic interaction terms Variable Coefficient Std. Err Time Invariant Characteristics Maternal Education 0.05 0.04 Maternal Education x other ethnicity 0.08 0.06 Paternal Education 0.05 0.04 Paternal Education x other ethnicity 0.04 0.05 Poverty Status (1979) 0.39 0.3 0 Poverty Status (1979) x other ethnicity 0.22 0.45 Library Card (1979) 0.09 0.28 Library Card (1979) x other ethnicity 0.36 0.43 HH Rec'd Newspaper (1979) 0.56* 0.3 0 HH Rec'd Newspaper (1979) x other ethnicity 1.24** 0.52 HH Rec'd Magazines (1979) 0.14 0.26 HH Rec'd Magazines (1979) x other ethnicity 0.08 0.39 Female 0.28 0.26 Female x other ethnicity 0.45 0.34 ASVAB percentile 0.02*** 0.01 ASVAB percentile x other ethnicity 0.002 0.01 Other Ethnicity 0.93 2.31 Time Variant Characteristics Former military status 0.07 0.37 Former military status x other ethnicity 0.13 0.48 Age 0.03 0.03 Age x other ethnicity 0.01 0.05 Experience 0.06 0.05 Experience x other ethnicity 0.04 0.07 Number of Children 0.1 0 0.14 Number of Children x other ethnicity 0.25 0.22 Health Status 0.23 0.38 Health Status x other ethnicity 0.34 0.42 Log Income 0.24 0.15 Log Income x other ethnicity 0.004 0.19 Out of Labor Force 0.02 0.01 Out of Labor Force x other ethnicity 0.01 0.01 Unemployed 0.04* 0.02 Unemployed x other ethnicity 0.02 0.03 Afr.Amer/Hisp subjects: 442 Afr.Amer/Hisp failures: 89 Other ethnicities subjects: 510 Other Ethnicities failures: 152 Notes: ***Significant at p<.01; **Significant at p<.05, *Significant at p<.10. Time variant characteristics lagged prior to final pre enrollment spell.

PAGE 134

123 Table 25 : Event history analysis with poverty interaction terms Variable Coefficient Std. Err Time Invariant Characteristics Maternal Education 0.004 0.04 Maternal Education x poverty status (1979) 0.02 0.08 Paternal Education 0.09*** 0.03 Paternal Education x poverty status (1979) 0.09 0.07 Poverty Status (1979) 1.74 3.07 Library Card (1979) 0.19 0.24 Library Card (1979) x poverty status (1979) 0.23 0.51 HH Rec'd Newspaper (1979) 0.16 0.26 HH Rec'd Newspaper (1979) x poverty status (1979) 0.24 0.59 HH Rec'd Magazines (1979) 0.11 0.21 HH Rec'd Magazines (1979) x poverty status (1979) 0.0003 0.5 0 Female 0.08 0.18 Female x poverty status (1979) 0.43 0.46 ASVAB percentile 0.02*** 0.005 ASVAB percentile x poverty status (1979) 0.001 0.01 Other ethnicity 0.05 0.2 0 Other ethnicity x poverty status (1979) 0.17 0.55 Time Variant Characteristics Former military status 0.01 0.33 Former military status x poverty status (1979) 0.86 0.76 Age 0.05* 0.03 Age x poverty status (1979) 0.06 0.06 Experience 0.03 0.04 Experience x poverty status (1979) 0.04 0.09 Number of Children 0.17 0.12 Number of Children x poverty status (1979) 0.13 0.28 Health Status 0.03 0.19 Health Status x poverty status (1979) 0.26 0.38 Log Income 0.2 0 ** 0.1 0 Log Income x poverty status (1979) 0.23 0.28 Out of Labor Force 0.01 0.01 Out of Labor Force x poverty status (1979) 0.01 0.02 Unemployed 0.02 0.01 Unemployed x poverty status (1979) 0.06 0.05 In poverty subjects: 198 In poverty failures: 33 Not in poverty subjects: 754 Not in poverty failures: 208 Notes: ***Significant at p<.01; **Significant at p<.05, *Significant at p<.10. Time variant characteristics lagged prior to final pre enrollment spell.

PAGE 135

124 Model Diagnostics Overall, a range of model diagnostics suggest s that this log normal specification is suitable for the available data. The likelihood ratio statistic shows that for each of the specified models above I can reject the null hypothesis that none of ate Additional diagnostics also support choosing the log normal specification over other potential distributions of parametric models. Using the Akaik e information criterion ( AIC) shows that the log normal distribution has the best fit from among the available options of parametric distributions (Mills, 2011) Another key diagnostic test involves examining the Cox Snell residuals (Bradburn, et al., 2003 ; Mills, 2011). In a properly fitted model, these residuals should have an exponentia l distribution with a slope of one (Box Steffensmeier & Jones, 2004; Bradburn, et al., 2003; Mills, 2011). Although plots of Cox Snell residuals show some deviation from a perfect slope of 1 in the models above it is limited to a relatively small number of observati ons. Overall, these residuals show a strong fit. Using a Cox proportional hazards model was another option for these estimations. However, post regression diagn ostics using this approach shows some limitations. Although the Cox Snell residuals and likelihood ratio tests suggest reasonable fit, further examination shows that these data would violate the proportional hazards assumption which is a necessary assumpt ion for employing this model (Mills, 2011; Orbe, Ferreira, & Nunez Anton, 2002) This test was conducted by examining the Schoe nfield residuals, which showed statistically sign ificant coefficients over time, suggesting that several of the covariates violate the proportional

PAGE 136

125 hazards assumption (Mills, 2011) Thus, I elected to forego using a Cox model and ch ose the log normal appro ach from among the available parametric models

PAGE 137

126 CHAPTER VI DISCUSSION & CONCLUSIONS The findings reported above provide mixed support for the four hypotheses guiding this research. There is strong evidence that some near completers who finish degrees receive a wage premium, but the return is not positive across all sub populations. The finding that some subgroups receive no statistically significant return, with others receiving high returns, partially confirms hypoth esis 1.1. This hypothesis states that near completers who finish a degree will earn a wage premium compared to those who do not However, t he results also confirm hypothesis 1.2, which states that the wage premiums of degree completion do not accrue to sub groups equally. With African Americans/Hispanics receiving statistically significant and substantively large wage premiums compared to insignificant premiums for other ethnicities, the data support the conclusion that premiums are not equal across these gr oups. Although the scenarios approximating the return to education use both the average population effect as well as the effect found specific to African Americans/Hispanics, the latter should carry more weight. Based on the analysis here, it would be faul ty to conclude that all near completers who return to finish a baccalaureate degree are likely to earn a positive economic return. The findin gs partially confirm hypothesis 2 which states that individuals graduating from public institutions would earn hi gher returns than those graduating from other institutions T he results are dependent on the different estimated costs associated wi th attendance at public, non profit and for profit schools. Caution is warranted due to the small number of individuals in the sample who graduated from for profit colleges (13 in total) and due to the assumptions made about actual tuition prices paid by individuals

PAGE 138

127 reenrolling in different sectors. With the general assumption that non profit and for profit tuition prices are higher, graduates from these sectors would either have to realize higher wage premiums or lower indirect costs (perhaps due to the flexibility of such programs allowing them to work more hours) in order to generate the same long term economic return as ind ividuals paying lower tuition at public schools. The event history analysis partially supports hypothesis 3, with one key predictor of traditional student success also being a predictor of near completers finishing degrees However, m any other factors that generally predict degree attainment do not appear to pre dict degree completion for near completers T hese findings are all discussed in greater detail below. Economic Returns to Degree C ompletion The economic return to degree completion by near completers depends on several factors: the wage premium for degree completion, direct and indirect costs, and the years an individual has remaining in the workforce. The discussion of results focuses first on the results from the wage premium estimations b efore turning to the fully calculated rates of return for degree completion. Wage premiums. The evidence presented above suggests that there is indeed a wage premium for near completers who finish degrees. T he first model, which includes the full sample w ithout interactions shows a general effect of about 1 7 percent higher earnings for near completers who finish degrees These results are somewhat sensitive to the appendix The model with gender interaction terms does not show any statistically

PAGE 139

128 significant differences (other than the association between earnings and the number of children). 31 The model with interaction terms for racial/ethnic background does show statistica lly significant and substantively large differences in the wage premium for degree completion. African Americans/Hispanics earn about a 31 percent premium for degree completion, while those of other ethnicities (again, who are likely mostly white) receive much smaller premiums. There are several potential explanations for these differential outcomes. It could be that the t ypes of jobs held or sought by African Americans/Hispanics who finish degrees are more responsive to the additional credential or that t he types of degrees pursued are different than other ethnicities and these degrees are more valuable in the workplace Alternatively, it could be that the degrees earned serve as a valuable signal, potentially offsetting racial/ethnic wage disparities Det ermining whether the higher return I find is due to choosing high wage fields, shifting from part time to full time work greater buffering against racial/ethnic wage disparities or some additional factor is a potential avenue for future research. As id entified in the literature review, studies that examine how racial/ethnic background interacts with degree completion have reached contradictory conclusions, with some showing higher returns for racial/ethnic minorities due to the sheepskin effect, and oth ers showing White individuals typically have higher returns (Belman & Heywood, 1991; Bitzan, 2009). Other research concludes that when examining labor market returns to high school diplomas, there are racial ly /ethnic ally based differences, such that African Americans receive lower wages than white counterparts of similar ability levels 31 This may indicate differences in child care responsibilities and is consistent with other literature in the field but is not necessarily releva nt to the study at hand.

PAGE 140

129 (Arcidiacono, Bayer, & Hizmo, 2010). In comparison, that same research shows that college diplomas level the playing field so that wages a re more related to ability (Arcidiacono, Bayer, & Hizmo, 2010). Evidence that African Americans/Hispanics receive a higher wage gain for degree completion would be consistent with these other conclusions In essence, finishing a degree could help racial/et hnic minorities move out of a labor market where there is evidence of race based wage disparities into a more equitable one. Heckman and Lafontaine (2006) also find that African Americans and Hispanics receive higher wage premiums for degree completion tha n whites, but they do not directly test the equality of the coefficients. Thus, the findings of higher wage premiums for racial/ethnic minorities are consistent with other research, but additional study to examine the mechanism that leads to differential p remiums specific to near completers is warranted. Additionally, further research that examines whether returns differ between African Americans and Hispanics (as well as other racial/ethnic groups) would be useful. Combining the two groups, which was nece ssitated by the data, could mask substantial differences between the two groups. There is some evidence in the literature on general returns to education that African Americans of all education levels earn less than Hispanics ( see for example Grubb, 1993). This type of comparison, however, does not examine whether the returns differ. In this vein of research, it could be that while African Americans earn less than Hispanics, they still could both receive equal benefits (as a percentage of their average earn ings) for attaining higher education levels. Few studies directly test differences in returns to education between these ethnicities and instead focus on comparisons to white individuals. Studies explicitly focused on differences by

PAGE 141

130 race appear not to find major differences in returns to education by racial/ethnic background ( Barrow & Rouse, 2005; Reimers, 1984). The estimations based on sector show no evidence that wage premiums are affected by the sector of the school from which a near completer graduates. The results are suggestive, with a higher coefficient for for profit graduates, but the results are not statistically significant. The fact that fewer than 15 near completers finished degrees at for profit schools in this sample suggests further study may be warranted with data that provide a larger sample size. In particular, more recent da ta that accounts for the growth in enrollment at for profit colleges and universities may be helpful in further analyzing returns by sector. Costs and returns to education Many of the estimates of the return to education cited in the literature review appear to overstate the economic return to degree completion for near completers by omitting direct and indirect costs. Research that considers direct and indir completers than those typically cited in the policy community (Abel & Deitz, 2014). T he results presented here suggest a lower economic return for near completers. Further, the return is hig hly dependent on the number of years an individual has remaining in the workforce. As shown th rough the series of plots in the previous section it can take many years for an investment in degree completion to turn positive. For policymakers and practitio ners, one key lesson is the importance of foregone wages in the overall cost model. The data show that for this cohort at least, going back to college reduces wages. The estimations here showed that these lost wages tend to be even larger than average tui tion at public colleges and still amount to a substantial portion of

PAGE 142

131 the total cost at for profit and private schools. In discussions of financial aid and college affordability, foregone wages are seldom mentioned, yet as can be seen here, they account for a substantial portion of the overall cost. As many of the near completers considered here are likely in the middle of careers with major financial obligations and responsibilities, the potential loss of wages may be a serious deterrent to returning to sch ool. Given that the coefficient for earnings while an individual is enrolled points to lower incomes, but does not suggest that returning near completers are giving up all of their incomes, it is likely that they are still on average working significant ho urs. Degree completion programs that provide greater flexibility, which have increased in recent years particularly through the growt h of distance education, could reduce this burden. Few of the degree completers from this cohort completed degrees in recent years, but analysis of a younger cohort where near completers are more likely to complete degrees through online education could analyze whether such programs are successful in reducing indirect costs. Colleges and universities could also examine th e impact of flexibility in non academic offerings as well, including providing student services, such as advising, financial aid assistance, child care options, and registrar related tasks through more flexible means or at non traditional times that would accommodate work schedules. Also, further research is warranted to understand better why African Americans/Hispanics have substantial positive re turns, while the predominantly white racial/ethnic group do es not. The next step in fully understanding how ne ar completers who finish degrees benefit is to examine the mechanisms by which the credentials do or do not affect earnings. It could be that for racia l/ethnic majorities that credentials are less

PAGE 143

132 important in determining earnings. Alternatively, it could be that Africa n Americans/Hispanics tend to cho o se to return to degrees that are highly relevant in the workfo rce, leading to higher premiums, or that baccalaureate degrees help erase some of the wage disparities due to racial/ethnic background that are ev ident at other educational levels. Contributions to the field. This paper attempts to fill important gaps in both the academic literature on returns to education and the differential effects of public and private provision of services. The study also aim s to better inform the policy community Although there are no extant studies with which to compare findings, it is reasonable to see how these results fit within the broader context of literat ure on returns to degrees. The wage premiums estimated here tend to be smaller than those found elsewhere in the literature, usually due to differences in the comparison group. Typically, ly a high school diploma finding wage premiums in the range of 40 50 percent (Grubb, 1997; Kane & Rouse, 1995; Marcotte, 2005). As for whether returns are dependent on the sector from which an individual graduates, this study suggests that there is no sta tistically significant difference in wage premiums, although caution is warranted due to sample sizes. In the policy realm, this challenges conventional wisdom that for profit degrees are of less value in the workforce than public or non profit ones. It ap pears that those who finish degrees at for profit schools are relatively well served, but this study does not address those who drop out short of degree completion, which is a significant concern to policymakers. Given that

PAGE 144

133 most of the near completers who finish degrees over the course of the survey did so before significant rises in for profit enrollments also suggests that further study is warranted on effects by sector. In the academic literature, t his finding adds to the broader, yet inconclusive, resea rch about the variance of wage premiums by sector of graduation However, the analyses that show how tuition costs can dramatically affect returns, particularly for those individuals who have a limited amount of time remaining in their workforce, suggest t hat overall returns to private and for profit degrees are highly likely to be lower absent a substantial wage premium for graduating from these sectors or significant tuition discounts and/or financial aid Additional research on whether the time it takes near completers to graduate varies by sector would add further precision to these estimates, which may be important given that some schools, particularly in the for profit sector, suggest that they offer ways to reduce the time it takes to earn a degree. Above all, t h ese results show that the estimates of wage returns for near completers who finish degrees that are typically used in the p olicy environment are likely overstated. holders show premiums as high as 70 percent for degree completion (Minnesota State Colleges and Un iversities, 2015). Similarly constructed analyses cite an additional $1 completion programs, which would necessitate a similar wage premium (Council on Postsecondary Education Commonwealth of Kentucky, 2008). While these estimates and marketing materials do not explicitly say that this is the wage premium for near

PAGE 145

134 completers who finish degrees, the inference is clear. Further, these analyses rarely account for direct and indir ect costs, which I have shown are substantial and have a large effect on how quickly an investment in further education pays off Perhaps more importantly, this research shows that some sub populations may not ever earn a positive economic return. This un dercut s the broader argument in favor of degree completion programs targeted at the general population. Identifying differences in the types of degrees earned and career paths of those who earn positive returns compared to those who do not would help polic ymakers better formulate strategies for ensuring that those who return to finish their education are likely to receive a benefit or are at least aware of the likely outcomes. Additionally progra ms targeting near completers could help reduce income inequal ity between racial/ethnic groups due to the large r returns earned by minorities. Why Do Near Completers R eturn? The series of event history analysis models that examine the factors that are associated with changes in produ ce s several noteworthy findings. Many of the factors that are key explanatory variables in the general lit erature on education attainment have no statistical ly significant association with returning to finish a degree. Other proxy variables for the educati on environment that the individual experienced as a child, including access to libraries and other written materials also have no statistical significance in these models. Among these covariates only ASVAB score s and paternal education show statistical si gnificance. The finding on the effect of a

PAGE 146

135 Among ne ar completers, individual ability (as measured by the ASVAB) se ems to trump family background as many of the factors that are considered important predictors of educational attainment are not significant here Of particular interest, family poverty status, which is generally a very important factor in overall education al attainment, does not have a statistically significant relationship with survival time. Additionally, the model that interacts variables with familial poverty status does not suggest that this initial poverty level impacts the decision to return to finish a degree. It appears that as individuals attain near completer status, poverty status from their youth is not an important predictor of whether they return to graduate. While the pat ernal education and ASVAB score s are statistically significant, it is also important to examine whether the effect size is substantively large as well. Doing so first requires an examination of the amount of time it typically takes a near completer to fini sh a degree. On average, it takes near completers 5.5 years to return for the enrollment spell that leads to graduation. An increase of one standard deviation in paternal education (3.9 years) would therefore be associated with a decrease in the time it ta kes to return of 1.6 years. I ncreasing an ASVAB score by one standard deviation ( 24 .8 points) would be associated with a substantial decrease in the tim e it takes to return and finish a degree (about 2.6 years ) Among time variant characteristics, income is the lone variable with strong statistical significance (p<.05). The association between income and degree completion points in the opposite direction as one might expect based on literature focused on the traditional education pipeline, where higher fam ily income s are a strong pr edictor of

PAGE 147

136 education attainment. Here, increased income is associated with an increase in the time it takes a near completer to return and finish his or her degree. An increase in one standard deviation of income ( about 1.2 log points ) is associated with a 54.2 percent longer time to earn a degree, which amounts to about three additional years. Conversely, a standard deviation decrease in earnings is associated with 16 percent shorter survival times (about .9 years) T his confirm s the phenomenon noted in the literature review where lower income levels lead individuals to pursue additional education and training. However, when combined with the findings from the fixed effects models about the absence of an Ashenfelter Dip, and evid ence that incomes tend to rise just prior to enrollment, this conclusion requires some nuance. It appears that lower average levels of income lead individuals to return, but they may also need a short term increase in earnings to enroll. Generally speaking the covariates for income, experience, and employment status when less likely he or she is to return to finish a degree. T his could reflect the higher opportuni ty cost of doing so similar to conclusions reached by Jepsen and Montgomery (2012), or it could suggest that those with lower incomes are more likely to seek out ways to bolster their earnings (Blanden, et al., 2012). Differential impacts While the results discussed above pro vide some useful insights into the factors that generally may be associated with higher likelihoods of returning to finish degrees examining the results of the models that include interaction terms show limited differences among key sub populations. Few variables in the interaction models show statistical significance. Different operational definitions of near

PAGE 148

137 completer resulted in few additional factors showing statistical significance. These are discussed in g reater detail in Appendix 1. Conclusion s These findings are potentially the beginning of a significant thread of research in human capital development. Showing that the economic return to degree completion by near completers depends not only on tuition pri ces paid, but also on racial/ethnic background suggests that general assumptions underlying the push to have more near completers finish degrees may be premature. The research presented here clearly shows that positive economic returns do not accrue equall y to all individuals, and that overall returns (as opposed to wage premiums ) are likely very much dependent on the type of school an individual chooses (mainly due to differences in direct costs rather than differences in earnings). Further, identifying th at factors such as paternal education, performance on high school aptitude tests, and income all appear to affect the likelihood of a near completer returning to finish his or her degree can lead to important policy conclusions. However, limited findings w hen examining results by gender, racial/ethnic background and familial poverty status suggest that further research is warranted These conclusions can help inform efforts to increase local, state, and national degree attainment rates and help meet future workforce needs. It may be a questionable approach to conclude a dissertation with a discussion of its limitations and the questions left unanswered, but given the limited extant research into near completers the findings presented here represent only a first step in a much broader research program into the impacts and drivers of degree completion by near completers and the individual and societal impacts of such degrees. While state leaders

PAGE 149

138 and other education and economic development stakeholders routi nely tout the necessity of incenting a greater percentage of this population to return and finish degrees, very little is understood about why they return or the economic and non economic returns that both indiv iduals and society may realize (and whether those returns vary by the type of school an individual chooses). As this is the initial research effort into the economic outcomes of this population, studies that attempt to replicate these results using other datasets, different cohorts, and diffe rent t ime periods are warranted. This research also uncovers a range of potential new questions that could be examined using this (or another similar) dataset. These questions could include further analyses of ter returning to finish his or her degree. It could be that near completers who finish degrees change jobs or occupations moving to new firms or industries where they are rewarded for their degree. Alternatively, they could be promoted within the same fir m and receive higher wages. Additional analyses could also examine the types of degrees that near completers are earning to show whether they may help fill potential shortages of educated labor in all fields or if their degrees tend to be limited to certai n programs. Further, it could be that the field in which an individual pursues a degree imp acts his or her economic return and that certain types of degrees hold value, while others do not. Finally, it may be that individuals in fields where credentials ar e necessary to earn higher wages are much more likely to return to finish degrees which could be a source of selection bias This question could also be addressed through further research on the connection between pre enrollment jobs and the decision to r eturn to finish a degree.

PAGE 150

139 Conducting these analyses could take advantage of data from the NLSY that include occupation codes and codes for the major or field of study. To examine why certain near completers who finish baccalaureate degrees receive wage pre miums, it would be possible to use models with wages as the dependent variable and terms that interact degree completion with occupation changes as explanatory variables. Sample sizes would likely decrease too rapidly to test whether particular occupation codes are associated with greater wage differences but it might be feasible to test whether degrees in certain industries are more valuable using industry codes provided by the NLSY. Additional analysis could compare whether near completers finishing bacc alaureate degrees are moving into occupations in which labor market projections show long term shortages. Finally, further research could test the value (in wage changes) for near completers by different types of baccalaureate degrees earned using data on program of study in the NLSY. Additional research may be warranted to examine whether near completers who go back to school use student loans to finance the direct costs. Financing the direct costs of returning to finish a degree over a longer term could h ave implications for overall economic returns and warrants further study. Although there are not usable student loan data in the 1979 NLSY survey, the 1997 cohort does provide more information on student loan levels and would be an appropriate data source for examining related questions. Research on near completers could also provide important insights for discussions be that near completers, who tend to be older than t raditional students, are less likely to

PAGE 151

140 leave a state after finishing their degrees. This could make them a more attractive compared to traditional students that may have higher rates of post graduation mobility I nvestments in these individuals, through scholarships and financial aid, could have longer term benefits for the state than investments in individuals who are more likely to migrate to another state. Through the restricted use NLSY dataset, it should be possible to examine the migration patterns of near completers who finish degrees com pared to traditional students. Conducting this analysis would likely necessitate another event history analysis. Moving away from the jurisdiction of concern would be the key dependent variable. The key comparison would examine differences in the likelihood of moving based on being either a near completer who finished a baccalaureate degree or a traditional degree completer. The jurisdiction of interest would either likely be a state, or given the recent interest in city focused policies to promote degree attainment a metropolitan area. The necessary geo location data are available from the NLSY restricted use dataset. Additionally, with the growth of distance learning i n recent years, and new educational offerings that award degrees based on demonstrated competencies rather than accumulated credits, it could be that the opportunity costs for near completers to finish degrees is decreasing. Flexible education offerings sh ould in theory reduce the economic sacrifices returning near completers face in deciding to return to school. This research would be highly beneficial to schools and colleges seeking to design degree programs that reduce opportunity costs for returning stu dents. Along with the growth in distance learning, enrollment in for profit colleges and universities has dramatically increased in recent years. Research using a more recent dataset, such as the NLSY 1997 cohort, would

PAGE 152

141 likely find a greater number of indi viduals completing degrees at these types of schools and could add further clarity to the finding here that there are no significant differences in earnings of graduates by sector. Another unexamined question in this dissertation is the potential impact of option benefits because they now have the possibility of pursuing additional degrees such as masters or professional degrees, that provide even greater financial be nefits. Research on other education attainment threshold suggests that one of the big economic benefits of high school completion, for example, is that it allows the individual to go on to college, with its substantially larger economic returns (Heckman L ochner, & Todd, 2006 ). Finally, this research leaves unexamined the effects of returning to school on near completers who return and attempt to complete a degree but fail to do so. For states and others pushing degree completion programs aimed generally a t near completers, it is important to consider the implications of those who may return and incur substantial direct and indirect expenses, but not receive financial gains necessary to offset these costs. Presumably, these individuals end up worse off econ omically as a result of returning but not finishing a degree. Such information would be important both for understanding the full impact of such a policy approach and for providing more complete information to those individuals considering whether to retur n to try to finish a degree. While these questions remain unanswered and deserve further investigation, the findings above present evidence to answer some of the key fundamental q uestions about near completers. These findings challenge some of the assumpt ions held within the policy

PAGE 153

142 community, while filling in a blank spot in the academic literature, and in doing so, raise additional questions for future research

PAGE 154

143 REFERENCES 55,000 Degrees. (2014). 55,000 Degrees. Retrieved from www.55000degrees.org Abdul Diverse Issues in Higher Education Retrieved from http://diverseeducation.com/article/ 16362/ Adult College Completion Network (n.d.). Marketing and communications r esources. Retrieved from http://adultcollegecompletion.org/mktingCommunications Allison, P. (2009). Fixed Effect Regression Models Thousand Oaks, CA: Sage Publications. Allison, P. (1982). Discrete time methods for the analysis of event histories. In S. L einhardt (Ed.), Sociological Methodology (Vol. 13, pp. 61 9 8). San Francisco: Jossey Bass. Angrist, J., & Pischke, J. (2009). companion Princeton: Princeton University Press. Apling, R. (1993). Proprietary sch ools and their students. The Journal of Higher Education 64 (4), 379 416. Arrow, K. (1973). Higher education as a filter. Journal of Public Economics 2 193 216. Ashenfelter, O. (1978). Estimating the effect of training programs on earnings. The Review of Economics and Statistics 60 (1), 47 57. Retrieved from http://www.jstor.org/stable/10.2307/1924332 Ashenfelter, O., Harmon, C., & Oosterbeek, H. (1999). A review of estimates of the schooling/earnings relationship, with tests for publication bias. Labour Economics 6 453 470. Retrieved from http://www.sciencedirect.com/science/article /pii/S092753719900041X Ashenfelter, O., & Krueger, A. (1994). Estimates of the economic return to schooling from a new sample of twins. American Economic Review (December), 1157 1173. Athreya, K., & Eberly, J. (2015). The college premium, college noncompletion, and human capital investment (No. WP 13 02R). Richmond, VA. Bailey, M. J., & Dynarski, S. M. (2011). Gains and gaps: Changing inequality in U.S. college entry and com pletion (No. 17633). Cambridge, MA. Bailey, T., Kienzl, G., & Marcotte, D. (2004). The return to a sub baccalaureate education: The effects of schooling, credentials and program of study on economic outcomes Washington, DC. Retrieved from http://academicc ommons. columbia.edu/catalog/ac:172592

PAGE 155

144 Baker, R., Bradburn, N., & Johnson, R. (1995). Computer assisted personal interviewing: An experimental evaluation of data quality and cost. Journal of Official Statistics1 11 (4), 413 431. Banzhaf, H., & Bhalla, G. ( 2012). Do households prefer small school districts? A natural experiment. Southern Economic Journal 78 (August 2009), 819 841. Retrieved from http://journal.southerneconomic.org/doi/abs/10.4284/0038 4038 78.3.819 Barrow, L., & Rouse, C. E. (2005). Do retur ns to schooling differ by race and ethnicity? Chicago: Federal Reserve Bank of Chicago. WP2005 02. Bastedo, M. N., & Flaster, A. (2014). Conceptual and methodological problems in research on college undermatch. Educational Researcher 43 (2), 93 99. doi: 10.3102/0013189X14523039 Baum, C. F. (2001). Residual diagnostics for cross section time series regression models. The Stata Journal 1 101 104. Baum, S., Ma, J., & Payea, K. (2013). Education Pays 2013 Education Pays 2013 New York. Retrieved from http ://trends.collegeboard.org/sites/default/files/education pays 2013 full report 022714.pdf Becker, G. (1993). Human capital: A theoretical and empirical analysis with special reference to education (3rd ed.). Chicago: University of Chicago Press. Behrman, J A test and new estimates. Economics of Education Review 18 159 167. Retrieved from http://www.sciencedirect.com/science/article/pii/S0272775798000338 Bellevue University. (2013) Degree completion at Bellevue University. Retrieved from http://www.bellevue.edu/ways to learn/degree completion programs.aspx Belman, D., & Heywood, J. S. (1991). Sheepskin effects in the returns to education: An examination of women and minorities. Th e Review of Economics and Statistics 720 724. Bitzan, J. D. (2009). Do sheepskin effects help explain racial earnings differences? Economics of Education Review 28 (6), 759 766. doi:10.1016/j.econedurev.2008.10.003 Bjrklund, A., & Kjellstrm, C. (2002) Estimating the return to investments in education: how useful is the standard Mincer equation? Economics of Education Review 21 195 210. Retrieved from http://www.sciencedirect.com/science/article /pii/S0272775701000036 Black, D. a., & Smith, J. a. (20 04). How robust is the evidence on the effects of college quality? Evidence from matching. Journal of Econometrics 121 (1 2), 99 124. doi:10.1016/j.jeconom.2003.10.006

PAGE 156

145 Blanden, J., Buscha, F., Sturgis, P., & Urwin, P. (2012). Measuring the earnings returns to lifelong learning in the UK. Economics of Education Review 31 (4), 501 514. doi:10.1016/j.econedurev.2011.12.009 Boshier, R., & Collins, J. B. (1985). The Houle Typology after twenty two years: A large scale empirical test. Adult Education Quarterly 3 5 (3), 113 130. doi:10.1177/0001848185035003001 Bound, J., & Krueger, A. B. (1991). The extent of measurement error in longitudinal earnings data: Do two wrongs make a right? Journal of Labor Economics 9 (1), 1. doi:10.1086/298256 Box Steffensmeier, J., & Jones, B. (2004). Event History Modeling: A Guide for Social Sciences New York: Cambridge University Press. Bozeman, B. (1987). All organizations are public: Bridging public and private organizational theories San Francisco: Jossey Bass. Bozeman, B. (2013). What organization theorists and public policy researchers can learn from one another: Publicness theory as a case in point. Organization Studies 34 (1998), 169 188. doi:10.1177/0170840612473549 Bozeman, B., & Bretschneider, S. (1994). T theory: A test of alternative explanations of differences between public and private organizations. 4 (2), 197 223. Retrieved from http://jpart.oxfordjournals.org/content /4/2/197.short Bradburn, M. J., Clark, T. G., Love, S. B., & Altman, D. G. (2003). Survival analysis Part III: multivariate data analysis -choosing a model and assessing its adequacy and fit. British Journal of Cancer 89 (April), 605 611. doi:10.1038/sj. bjc.6601120 Brand, J. E., & Xie, Y. (2010). Who benefits most from college? Evidence for negative selection in heterogeneous economic returns to higher education. American Sociological Review 75 (2), 273 302. doi:10.1177/0003122410363567 Brewer, D., Eide, E., & Ehrenberg, R. (1999). Does it pay to attend an elite private college? Cross cohort evidence on the effects of college type on earnings. Journal of Human Resources 34 (1), 104 123. Retrieved from http://www.jstor.org/stable/146304 Bureau of Labor Sta tistics. (n.d.). The NLSY79 sample: An introduction. Retrieved from https://www.nlsinfo.org/content/cohorts/nlsy79/intro to the sample/nlsy79 sample introduction Card, D. (1999). The causal effect of education on earnings. Handbook of Labor Economics 3 1801 1863. Retrieved from http://www.sciencedirect.com/science/ article/pii/S1573446399030114

PAGE 157

146 Carneiro, P., Hansen, K., & Heckman, J. (2003). Estimating distributions of treatment effects with an application to the returns to schooling and measurement of t he effects of uncertainty on college. International Economic Review (44), 361 422. Carnevale, A., Smith, N., & Strohl, J. (2010). Help wanted: Projections of job and education requirements through 2018 Washington, DC. Cellini, S. R. (2012). For profit hi gher education: An assessment of costs and benefits. National Tax Journal 65 (March), 153 180. Cellini, S. R., & Chaudhary, L. (2014). The labor market returns to a for profit college education. Economics of Education Review 43 125 140. Chevalier, A., income and education on the schooling of their children. IZA Journal of Labor Economics 2 (8), 1 22. doi:10.1186/2193 8997 2 8 College Board. (2013). Average tuition and fee and room a nd board charges 1974 2013. Retrieved from http://trends.collegeboard.org/college pricing/figures tables/tuition and fee and room and board charges over time 1973 74 through 2013 14 selected years Council on Postsecondary Education, Commonwealth of Kentuck y. (2008). http://knowhow2goky.org/pg/adults_pg_ why.php Connecticut House Bill 5050. (2014 ). An act improving college completions. https://www.cga.ct.gov/2014/TOB/H/2014HB 05050 R00 HB.htm Culver, S. M. (1993). A survey of adult degree program alumni and current students at one university. The Journal of Continuing Higher Education 41 (2), 23 44. Dahl, R., & Lindblom, C. E. (2000). Politics, Economics, and Welfare (2nd ed.). New Brunswick, NJ: Tran saction Publishers. Dale, S., & Krueger, A. (2002). Estimating the payoff to attending a more selective college: An application of selection on observables and unobservables. Quarterly Journal of Economics (November), 1491 1527. Retrieved from http://www. nber.org/papers/w7322 Day, J. C., & Newburger, E. (2002). The big payoff: Educational attainment and synthetic estimates of work life earnings Washington, DC: United States Census Bureau. Deaton, A. (2010). Instruments, randomization, and learning about d evelopment. Journal of Economic Literature 48 (2), 424 455. Retrieved from http://www.jstor.org/stable /10.2307/20778731

PAGE 158

147 Deming, D., Goldin, C., & Katz, L. (2011). The for profit postsecondary school sector: Nimble critters or agile predators? National Bureau of Economic Research Working Paper Series Retrieved from http://www.nber.org/papers/w17710 Drukker, D. (2003). Testing for serial correlation in linear panel data models. The Stata Journal 3 (2), 168 177. Duderstadt, J., & Womack, F. (2003). The fu ture of the university in America: Beyond the crossroads Baltimore: Johns Hopkins University Press. EducationDynamics. 2010. Attracting adults back to college: A survey of WV adults. by adult workers. Provided to author via email. Ehrenberg, R. G. (2004). Econometric studies of higher education. Journal of Econometrics 121 (1 2), 19 37. doi:10.1016/j.jeconom.2003.10.008 The race is to the swift: Socioeconomic origins, adult education, and wage attainment. American Journal of Sociology 110 (1), 123 160. Retrieved from http://www.jstor.org/stable/10.1086/386273 d entry into work related education and training among adult workers. Social Science Research 31 (1), 49 76. doi:10.1006/ssre.2001.0718 Ewert, S. (2013). Expanded measures of education and their labor market outcomes. In Annual Meeting of the Population As sociation of America New Orleans, LA. Evans, W. N., Murray, S. E., & Schwab, R. (1997). Schoolhouses, courthouses, and statehouses after Serrano. Journal of Policy Analysis & Management 16 (1), 10 31. Feeney, M., & Welch, E. (2012). Realized publicness at public and private research universities. Public Administration Review 72 (2 ), 272 284. doi:10.111/j .1540 6210.2011.02521.x.Although Gangl, M. (2006). Scar effects of unemployment: An assessment of institutional complementarities. American Sociological Re view 71 (6), 986 1013. Retrieved from http://asr.sagepub.com/content/71/6/986.short Gibson, J. B. (2011). Organizational performance: The case of public ver sus private higher education. Remarks to the annual proceedings of the Association for Education Fin ance and Policy Government Accountability Office. (2005). Postsecondary institutions could promote more consistent consideration of coursework by not basing determinations on accreditation Washington, DC.

PAGE 159

148 Green, G., Ballard, G., & Kern, D. (2007). Return on investment: Assessing a nontraditional, interdisciplinary degree and career impact. The Journal of Continuing Higher Education 55 (1), 16 26. doi:10.1080/07377366.2007.10400105 Griliches, Z. (1977). Estimating the returns to schooling: Some econometric problems. Econometrica: Journal of the Econometric Society 45 (1), 1 22. Retrieved from http://www.jstor.org/stable/10.2307/1913285 Grubb, W. (1993). The varied economic returns to postsecondary education: New evidence from the class of 1972. Journal of Human Resources 28 (2), 365 382. Retrieved from http://www.jstor.org/stable/10.2307/146208 Hllsten, M. (2012). Is it ever too late to study? The economic returns on late tertiary degrees in Sweden. Economics of Education Review 31 (1), 179 194. doi:10.101 6/j.econedurev.2011.11.001 Hamilton, L. (2009). Statistics with Stata (2nd ed.). Belmont: Brooks/Cole. Harris, B. (2003). Application of curriculum learning outcomes from an adult baccalaureate degree completion program. The Journal of Continuing Higher Ed ucation (December 2013), 37 41. Retrieved from http://www.tandfonline.com/ doi/pdf/10.1080/07377366.2003.10400253 Hausman, J. (1978). Specification tests in econometrics. Econometrica 46 (6), 1251 1271. doi:10.2307/1913827 Hausman, J., & Taylor, W. (1981 ). Panel data and unobservable individual effects. Econometrica: Journal of the Econometric Society 49 (6), 1377 1398. Retrieved from http://www.jstor.org/stable/10.2307/1911406 HCM Strategists. (2014). States with higher education attainment goals Washin gton, DC. Heckman, J. J., & Hotz, V. J. (1989). Choosing among alternative nonexperimental Journal of the American Statistical Association 84 (408), 862 874. Heckman, J. Lochner, L., & Todd, P. (2006). Earnings functions, rates of return and treatment effects: The Mincer equation and beyond. Handbook of the Economics of Education 1 (06), 307 458. doi:10.1016/S1574 0692(06)01007 5 Heckman, J. J., & Lafontaine, P. A. (2006 ). Bias corrected estimates of GED returns (National Bureau of Economic Research Working Paper No. 12018). Cambridge, MA. Heckman, J. Lochner, L., & Todd, P. (2008). Earnings functions and rates of return IZA Discussion Papers No. 3310

PAGE 160

149 Heckman, J., & S mith, J. (1999). The pre programme earnings dip and the determinants of participation in a social programme. Implications for simple programme evaluation strategies. The Economic Journal 109 (457), 313 348. Retrieved from http://onlinelibrary.wiley.com/doi /10.1111/1468 0297.00451/abstract Hernandez, D. (1999). Comparing response rates for SPD, PSID, and NLSY Washington, DC: United States Census Bureau. Holzer, H., & Dunlop, E. (2013). ostsecondary education and labor market outcomes in the US IZA Discussion Papers No. 7319. Retrieved from http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2250297 Houle, C. (1961). The inquiring mind: A study of the adult who continues to learn. Madison, Wisconsin: University of Wisconsin Press Hout, M. (2012). Social and economic returns to college education in the United States. Annual Review of Sociology 38 (1), 379 400. doi:10.1146/annurev.soc.012809. 102503 Hout, M., Levanon, A., & Cumberworth, E. (2011). Job loss and unemployment. In D. Grusky B. Western, & C. Wimer (Eds.), The Great Recession New York: The Russell Sage Foundation. achieving, low income students. National Bureau of Economic Research Working Paper Series Retrieved from http://www.nber.org/papers/w18586 Hoxby, C., & Turner, S. (2013). Expanding college opportunities for high achieving, low income students. Stanford Institute for Economic Policy Research Discussion Paper Retrieved from http://www8. gsb.columbia.edu/programs admissions/sites/programs admissions/files/finance/Applied Microeconomics/Caroline Hoxby.pdf Hoyt, J. E., & Allr ed, E. (2008). Educational and employment outcomes of a degree completion p rogram. The Journal of Continuing Higher Ed ucation 56 (2), 26 33. doi:10.1080/07377366.2008.10400150 Hungerford, T., & Solon, G. (1987). Sheepskin effects in the returns to education. The Review of Economics and Statistics 69 (1), 175 177. Retrieved from http://www.jstor.org/stable/10.2307/1937919 Hvidman, U., & Anderse n, S. C. (2013). The Impact of performance m anag ement in public and private o rganizations. Journal of Public Administration Research & Theory doi:10.1093/jopart/mut019 Jacobs, D. (2013). Public or private college. Is the outcome any different? October: Forbes

PAGE 161

150 Jaeger, D., & Page, M. (1996). Degrees matter: New evidence on sheepskin effects in the returns to education. The Review of Economics and Statistics 78 (4), 733 740. Retrieved from http://www.jstor.org/stable/10.2307/2109960 Jen Social and Economic Research, University of Essex, Colchester. Jepsen, C., & Montgomery, M. (2012). Back to school: An application of human capital theory for mature workers. Economics of Education Review 31 (1), 168 178. doi:10.1016/j.econedurev.2011.10.005 Jepsen, C., Troske, K., & Coomes, P. (2014 ). The labor market returns to community college degrees, diplomas, and certificates. Journal of Labor Economics 32 (1), 95 121. doi:10.1086/671809 Joy, L. (2003). Salaries of recent male and female college graduates: Educational and labor market effects. Industrial and Labor Relations Review 56 (4), 606 621. Retrieved from http://www.jstor.org/stable/3590959 Julian, T., & Kominski, R. (2011). Education and synthetic work life earnings estimates Wash ington, DC: United States Census Bureau. Kane, T., & Rouse, C. (1995). Labor market returns to two and four year college. The American Economic Review 85 (3), 600 614. Retrieved from http://www.jstor.org/stable/10.2307/2118190 Kane, T. J., Rouse, C., & Staiger, D. (1999). Estimating returns to schooling when schooling is misreported (No. Working Paper 7235). Cambridge, MA. Kentucky Council on Postsecondary Education. (2008). Project Gr aduate. Retrieved from http://www.knowhow2goky.org/adults/adults_goback_why.php Kilpi Jakonen, E., de Vilhena, D. V., Kosyakova, Y., Stenberg, A., & Blossfeld, H. P. (2012). The impact of formal adult education on the likelihood of being employed: A compar ative overview. Studies of Transition States and Socieities 4 (1), 48 68. Retrieved from http://www.ceeol.com/aspx/getdocument.aspx?logid=5&id=45e1fe20dccd 4cba82ae5aa5fe063e3a Knott, J. & Payne, A. (2004). The impact of state governance structures on mana gement and performance of public organizations: A study of higher education institutions. Journal of Policy Analysis and Management 23 (1), 13 30. doi:10.1002/pam.10176 Lane, P., Michelau, D., & Palmer, I. (2012). Going the distance in adult college comple tion: Lessons from the Non traditional No More project. Boulder, CO: Western Interstate Commission for Higher Education. Retrieved from: w ww.wiche.edu/info /publications/ntnmStateCaseStudies.pdf

PAGE 162

151 Lang, K., & Weinstein, R. (2013). The wage effects of not for profit and for profit certifications: Better data, somewhat different results. Labour Economics 24 230 243. doi:10.1016/j.labeco.2013.09.001 Leigh, D., & Gill, A. (1997). Labor market returns to community colleges: Evidence for returning adults. Journal of Human Resources 32 (2), 334 353. Retrieved from http://www.jstor.org/stable/146218 Light, A. (1995). The effects of interrupted schooling on wages. Journal of Human Resources 30 (3), 472 502. Retrieved from http://www.jstor.org/stable/10.2307/146032 Li ght, A. (1996). Hazard model estimates of the decision to reenroll in school. Labour Economics 2 381 406. Retrieved from http://www.sciencedirect.com/science /article/pii/092753719580042V Light, A., & Strayer, W. (2004). Who receives the college wage premium? Assessing the labor market returns to degrees and college transfer patterns. Journal of Human Resources 39 (3), 746 773. Retrieved from http://jhr.uwpress.org/content/XXXIX/3/ 746.short Lumina Foundation. (2013). A s tronger nation through higher education. Indianapolis, IN. Retrieved from http://www.luminafoundation.org/publications/A_stronger_nation_ through_higher_education 2013.pdf Lumina Foundation. (2011). Strategic plan: 2011. Retrieved from http ://www.luminafoundation. org/wp content/uploads/2011/02/Lumina_Strategic _Plan.pdf MaCurdy, T., Mroz, T., & Gritz, R. (1998). An evaluation of the national longitudinal survey on youth. The Journal of Human Resources 33 (2), 345 436. Retrieved from http://w ww.jstor.org/stable/10.2307/146435 Maguire and Associates. 2010. College Access Challenge Grant research, summary of key findings and recommendations. Provided to author via email. Marcotte, D., Bailey, T., Borkoski, C., & Kienzl, G. (2005). The returns of a community college education: Evidence from the National Education Longitudinal Survey. Educational Evaluation and Policy Analysis 27 (2), 157 175. doi:10.3102/01623737027002157 Marcus, R. (1986). Earnings and the decision to return to school. Economics of Education Review 5 (3), 309 317. Retrieved from http://www.sciencedirect.com/science/article/ pii/0272775786900828 Maryland Senate Bill 0740. (2013). Maryland college and career readiness and college completion act of 2013. http://mgaleg.maryland.gov/we bmga/frmMain.aspx?pid= billpage&stab=01&id=sb0740&tab=subject3&ys=2013rs

PAGE 163

152 McKinney, W. (1991). Graduates' satisfaction with bachelor of general studies degree. J ournal of Continuing Higher Education 39(1), 16 18. ). Comparing the impact of public and private sector management: A preliminary analysis using colleges and universities. In Annual Meeting of the American Political Science Association Toronto. Retrieved from http://papers.ssrn.com/Sol3/papers.cfm?abstrac t_id=1450678 Mills, M. (2011). Introducing survival and event history analysis London: Sage Publications. Mincer, J. (1974). Schooling, Experience, and Earnings (Vol. I). New York: Columbia University Press. Retrieved from http://www.nber.org/books/minc74 1 Mincer, J. (1958). Investment in human capital and personal income distribution. Journal of Political Economy 66 (4), 281 302. Minnesota State Colleges and Universi ties. 2015. Graduate Minnesota: Complete your degree anytime. Anywhere. http://www.mnscu.edu/graduateminnesota/ Minnesota State Colleges and Universities. 2013. Graduate Minnesota Update. Provided to author via email. Minnesota State Colleges and Universit ies. (2015 ). Why return to college? Retrieved from h ttp://www.mnscu.edu/graduateminnesota/why.html Research in Higher Education 19 (2), 213 230. Retrieved from http://link.spring er.com/article/ 10.1007/BF00974760 Mishler, C, management: Theoretical expectations. Journal of Public Administration Research and Theory 21 (Supplement 3), i283 i299. doi:10.1093/jopart/m ur027 Monks, J. (2000). The returns to individual and college characteristics: Evidence from the National Longitudinal Survey of Youth. Economics of Education Review 19 279 289. Retrieved from http://www.sciencedirect.com/science/article/pii/ S0272775799 000230 Moore, H. (1970). Laws of Wages (2nd Editio n .). New York: Augustus M. Kelley. Morstain, B., & Smart, J. (1974). Reasons for participation in adult education courses: A multivariate analysis of group differences. Adult Education Quarterly 24 (2), 83 98. doi:10.1177/074171367402400201 Moulton, S. (2009). Putting together the publicness puzzle: A framework for realized publicness. Public Administration Review 889 900. Retrieved from http://onlinelibrary.wiley.com/doi/10.1111/j.1540 6210.2009.02038.x/full

PAGE 164

153 Murphy, K., & Welch, F. (1990). Empirical age earnings profiles. Journal of Labor Economics 8 (2), 202 229. National Center for Education Statistics. (2014). Digest of Education Stati stics Table 224. Retrieved from http://nces.ed.gov/programs/digest/d12/tables/dt12_224.asp National Center for Education Statistics (2013). Digest of Education Statistics, Table 381. Retrieved from http://nces.ed.gov/programs/digest/d12/tables/dt12_381.a sp National Center for Education Statistics. (n.d.). Integrated Postsecondary Education Data System Glossary. Retrieved from: http://nces.ed.gov/ipeds/glossary/?charindex=G National Center for Higher Education Management Systems and Delta Project on Coll ege Costs. (2011). Closing the Degree Gap Interactive Model. Retrieved from http://www.adultcollegecompletion.org/content/degreeGapModel National Student Clearinghouse (2012 a ). Current Term Enrollment Report Fall 2012. Retrieved from http://nscresearchce nter.org/currenttermenrollmentestimate fall2012/ National Student Clearinghouse (2012 b ). Snapshot Report Adult Learners. Retrieved from http://nscresearchcenter.org/wp content/uploads/SnapshotReport4_Adult_ Learners.pdf Oklahoma State Regents for Higher Education. (n.d.). Reach Higher. Retrieved December 1, 2013 from http://www.okhighered.org/reachhigher/ Orbe, J., Ferreira, E., & Nez Antn, V. (2002). Comparing proportional hazards and accelerated failure time models for survival analysis. Statistics i n Medicine 21 (January 2001), 3493 3510. doi:10.1002/sim.1251 Park, J. (1999). Estimation of sheepskin effects using the old and the new measures of educational attainment in the Current Population Survey. Economics Letters 62 237 240. Retrieved from htt p://www.sciencedirect.com/science/article/pii /S0165176598002262 Perry, J., & Rainey, H. (1988). The public private distinction in organization theory: A critique and research strategy. Academy of Management Review 13 (2), 182 201. Retrieved from http://am r.aom.org/content/13/2/182.short Rainey, H., & Bozeman, B. (2000). Comparing Public and Private Organizations: Empirical Research and the Power of the A Priori. Journal of Public Administration Research and Theory 10 (2), 447 470. doi:10.1093/oxfordjournal s.jpart.a024276 Rainey, H., Backoff, R., & Levine, C. (1976). Comparing public and private organizations. Public Administration Review 36 (2), 233 244. Retrieved from http://www.jstor.org/stable/10.2307/975145

PAGE 165

154 Reimers, C. (1984). The wage structure of Hisp anic men: implications for policy. Social Science Quarterly 65 (2), 401 16. Schlesinger, R. (2010, August). Is a $50,000 college tuition worth it? U.S. News and World Report Retrieved from http://www.usnews.com/opinion/articles/2010/08/17/is a 50000 coll ege tuition worth it Schoeni, R. F., Stafford, F., McGonagle, K. a, & Andreski, P. (2013). Response rates in national panel surveys. The Annals of the American Academy of Political and Social Science 645 (1), 60 87. doi:10.1177/0002716212456363 Schultz, T. (1963). The Economic Value of Education. New York: Columbia University Press. Scott, M., Bailey, T., & Kienzl, G. (2006). Relative success? Determinants of college graduation rates in public and private colleges in the US. Research in Higher Education 47 (3), 249 279. doi:10.1007/sl Sen, B. (2006). Frequency of family dinner and adolescent body weight status: evidence from the national longitudinal survey of youth, 1997. Obesity 14 (12), 2266 76. doi:10.1038/oby.2006.266 Shapiro, D., Dundar, A., Yuan, X., Harrell, A., Wild, J., Ziskin, M. (2014). Some College, No Degree: A National View of Students with Some College Enrollment, but No Completion (Signature Report No. 7). Herndon, VA: National Student Clearinghouse Research Center. Simon, H. (1957). Administrative Behavior (2nd ed.). New York: The Macmillan Company. Skalli, A. (2007). Are successive investments in education equally worthwhile? Endogenous schooling decisions and non linearities in the earnings schooling relat ionship. Economics of Education Review 26 (2), 215 231. doi:10.1016/j.econedurev.2005.07.004 Solon, G., Haider, S. J., & Wooldridge, J. M. (2015). What are we weighting for? The Journal of Human Resources 50 (2), 301 316. Retrieved from http://www.nber.org / papers/w18859 Staehle, H. (1943). Ability, wages, and income. The Review of Economic Statistics 25 (1), 77 87. Retrieved from http://www.jstor.org/stable/1924549 Taniguchi, H., & Kaufman, G. (2005). Degree completion among nontraditional college students Social Science Quarterly 86 (4), 912 927. doi:10.1111/j.0038 4941.2005.00363.x Taylor, F. (1967). Principles of Scientific Management New York: W. W. Norton & Company.

PAGE 166

155 Tierney, M. (1980). The impact of financial aid on student demand for public/private higher education. The Journal of Higher Education 51 (5), 527 545. Retrieved from http://www.jstor.org/stable/1981405 Tourangeau, R., & Smith, T. W. (1996). Asking sensitive questions: The impact of data collection mode, question formate, and question cont ext. Public Opinion Quarterly 60 275 304. doi:10.1086/297751 U.S. Census Bureau. (n.d.). Historical poverty tables People. https://www.census.gov/hhes/www/poverty/data/historical/people.html U.S. Census Bureau. (2011). American Community Survey: Educational Attainment Retrieved from http://factfinder2.census.gov/faces/nav/jsf/pages/searchresults.xhtml ?refresh=t U.S. Department of Education. (2015). Fact sheet: Obama administration increases accountability for low performing for profit institutions. Retrieved from http://www.ed.gov/ne ws/press releases/fact sheet obama administration increases accountability low performing profit institutions U.S. Department of Education. (2011). 2011 College Completion Toolkit Retrieved from http://www.whitehouse.gov/sites/default/files/college_completion_tool_kit.pdf U.S. Government Printing Office. (2010). The Federal Investment in For Profit Education. Are Students Succeeding? Senate hearing transcript 111 11 62. U.S. Senate Health, Education, Labor, and Pensions Committee. (2012). For profit Education: The Failure to Safeguard the Federal Investment and Ensure Student Success. Staff report. University of Phoenix. (2015). Prior Learning Assessment. Retrieved fr om http://www.phoenix.edu/admissions/prior_learning_assessment.html Walmsley, G. L., & Zald, M. N. (1973). The political economy of public organizations Lexington, MA: Heath. Weber, M. (1970). Bureaucracy. In H. Gerth (Ed.), Max Weber: Essays in Sociolog y Oxford: Oxford University Press. Weiss, A. (1995). Human capital vs. signalling explanations of wages. Journal of Economic Perspectives 9 (4), 133 154. doi:10.1257/jep.9.4.133 White House. (n.d.). Higher education Retrieved December 5, 2013 from http:/ /www.whitehouse.gov/issues/education/higher education. Winship, C., & Radbill, L. (1994). Sampling weights and regression analysis. Sociological Methods and Research 23 (2), 230 257. doi:0803973233

PAGE 167

156 Wooldridge, J. (2006). Introductory econometrics: A modern approach Mason, OH: Thomson/South Wester n Wooldridge, J. (2002). Econometric analysis of cross section and panel data Cambridge: MIT Press. doi:10.1515/humr.2003.021 Zhang, L. (2005). Do measures of college quality matter? The effect of college quality on The Review of Higher Education 28 (4), 571 596. doi:10.1353/rhe.2005.0053

PAGE 168

157 APPENDIX Partly because there is no general consensus on research methodologies for examining economic returns to near completers, several key choices abo ut definitions, measurement, and estimation techniques were made throughout this study. Given the novelty of the subject, additional discussion about these approaches and the implications they have for the results is warranted. This appendix shows the implications of definitional and methodological choices made throughout the study by comparing results generated when make alternative choices. Where results show substantive differences, additional discussion is included. The appendix begins with consider ing different definitions of near completers, then shows how those definitions would affect the outcomes estimated from the fixed effects models and event history analyses. I then include two tests that show how the wage premium changes over time. I also s how how different assumptions about enrollment intensity could affect the length of time it takes a near completer to finish a baccalaureate degree resulting in differences in the approximated return. I then include a short section on the decision not to w eight the regression analyses and compare my results with a weighted regression. Definitions of Near Completer calaureate degree, then had a spell of non enrollment. Table 26 lists four other potential definitions. These definitions are integral to both the fixed effects models and the event history analyses presented above. The sections show how the outcomes of th e study would differ depending on the definition chosen. Table 26 : Definitions of near completer Definition 1 (in use) Completed at least 50 percent of a baccalaureate degree, then had a spell of non enrollment. Definition 2 Completed at least 50 percent of a baccalaureate degree, then made no further progress for at least three years (or two survey spells post 1994). Definition 3 Completed at least 50 percent of a baccalaureate degree, then had a spell of non enrollment of at least two years. Definition 4 Completed at least 75 percent of a baccalaureate degree, then had a spell of non enrollment. Sensitivity tests for wage premiums. In the models estimating the wage premiums for degree completion, there are several important methodological decisions that impact findings. To examine the differences that would result from using these definitions, I employ them in the fixed effects models that were used to estimate the return for degree completion, as well as interaction effects for gender, racial/ethnic background, and familial poverty status. The comparisons for the first model, which examines the average retur n across the entire subsample are presented in Table 27 below.

PAGE 169

158 Table 27 : Regressions with different definitions of near completer Variable Def. 1 Def. 2 Def. 3 Def. 4 Year prior to enrollment a 0.07** 0.07 ** 0.04 0.03 Enrolled a 0.17*** 0.14** 0.12*** 0.06 Graduate a 0.16*** 0.14** 0.1 0 0.1 0 Age 0.12*** 0.08** 0.1 0 *** 0.1 0 Age 2 0.001*** 0.001** 0.001* 0.001*** Exp 0.29*** 0.23 ** 0.25*** 0.32*** Exp 2 0.03*** 0.02 ** 0.02*** 0.03*** Exp 3 0.001*** 0.001** 0.001** 0.002*** Exp 4 0.00002*** 0. 00001 * 0.00001 0.00003** Wks unemployed 0.03*** 0.03** 0.03*** 0.03*** Wks out of labor force 0.03*** 0.03 ** 0.03*** 0.03*** Health limitations 0.02 0.03 0.02 0.04 Number of children 0.02 0.02 0.03 0.02 Never married b 0.08 0.03 0.07 0.17 Married b 0.14 0.05 0.09 0.14 Separated b 0.12 0.003 0.05 0.19 Divorced b 0.13 0.04 0.08 0.16 Observations 12,156 10,611 8,827 4,400 Individuals 1,042 958 786 389 Note: Year fixed effects also included but not reported. Significant at p<.1; ** Significant at p<.05; *** Significant at p<.01 a Reference category is near completers who are not enrolled, have not finished a degree, and do not enroll the following yea r. b Reference category is widowed The different definitions of near completer produce slightly different results. Definition 2 does not appear to lead to different findings compared to the definition used in this study (Definition 1). This definition leads to a slightly different subsample, as it did not consider spells of non enrollment. Although the intention of the definition is to serve as a proxy for individuals who have left postsecondary education, it captures individuals who are enrolled part t ime and progressing slowly towards a degree. Definition 3 examines slightly longer spells of non enrollment and appears to produce substantively different results, as the coefficient for graduates is not statistically significant. It is just outside of m oderate significance at p=.12. While this does suggest caution is warranted before concluding that these are vastly different findings, it could be that this indicates that the return to degree completion for near completers drops the longer they have been away. However, this definition only captures 156 near completers who finished baccalaureate degrees (compared to 319 under the definition in use), so shrinking sample sizes could be another issue. Definition 4, which includes individuals

PAGE 170

159 who completed 75 percent of a degree then had a spell of non enrollment shows similar results to the definition in use, but the statistical significance declines for the key term to only a modest level (p<.10). Again, this more restrictive definition captures fewer individ uals, so the size of the subsample could be a concern. Interaction Effects With the range of models employed to test for differential effects by sub group, it is also appropriate to test the different definitions above in the interaction models for gender, racial/ethnic background, and familial poverty levels. Table 28 shows that the choice of definition modest implications for results based on gender mainly because few of the interaction terms show any statistical significance. In the gender interaction models, across the interaction terms, the results are relatively robust, exc ept for the term for enrollment. Definitions 2 and 4 show higher earnings for women while enrolled compared to men with modest statistical significance (p<.10), while the other two definitions do not. Given that it is not strong statistical significance, i t is a relatively sound conclusion that the definitional choice for near completer does not affect the results for gender interactions. The coefficient for additional children for women also shows statistical significance across all four definitions, altho ugh only in the first two definitions is the base term significant. These are robust results showing that the The main finding from the racial/ethnic group interaction model is that individu als of African Americans/Hispanic race/ethnicity earn a higher return than the predominantly white group. Comparisons of the interaction model using different definitions for near completer are presented in the Table 29 below. The differences in the wage p remium across race/ethnic groups holds up across three of the four definitions. When using Definition 2, the difference is not statistically significant. This is the only definition that does not incorporate a spell of non enrollment, so it may be that the re are differences in how race/ethnicity affects graduates by their enrollment status. It could be that a diploma is a more important signal for those African Americans/Hispanics who have completely stepped away from postsecondary education rather than jus t slowed their progress. Alternatively, it could be that those African Americans/Hispanics who have left pursue a degree that is highly likely to yield positive wage premiums when they return. Although the familial poverty status interaction model does not produce statistically significant results as initially estimated, I still compare results across the different definitions of near completer. The results, presented in Table 30 show some sensitivity between the definitions for the education variables, bu t it is limited to the base terms. None of the interaction terms for education and poverty status show any statistical significance. Employing different definitions for near completer in the analyses of sector interaction sector effects similarly shows fe w statistically significant differences between the key education coefficients, as can been seen in Table 31 below.

PAGE 171

160 Table 28 : Differential effects by gender Variable Def. 1 Def. 2 Def. 3 Def. 4 Pre enrollment 0.09* 0.08* 0.03 0.1 0 Enrolled 0.2 0 *** 0.21*** 0.11* 0.2 0 ** Graduate 0.17*** 0.1 0 0.08 0.01 Pre enrollment x female 0.06 0.01 0.03 0.1 0 Enrolled x female 0.06 0.11* 0.06 0.18* Graduate x female 0.06 0.05 0.03 0.16 Age 0.14*** 0.12*** 0.14*** 0.15** Age 2 0.002*** 0.001*** 0.001*** 0.002*** Age x female 0.05 0.05 0.07 0.1 0 Age 2 x female 0.001 0.001 0.001 0.001 Exp 0.22*** 0.15*** 0.142** 0.24*** Exp 2 0.02*** 0.01 0.00864 0.03** Exp 3 0.001*** 0.0004 0.0003 0.001* Exp 4 0.00002** 0.000004 0.000003 0.00002 Exp x female 0.12 0.11 0.16** 0.16 Exp 2 x female 0.01 0.01 0.01 0.01 Exp 3 x female 0.001 0.0004 0.001 0.001 Exp 4 x female 0.00001 0.00001 0.00001 0.00001 Wks unemployed 0.03*** 0.03*** 0.03*** 0.03*** Wks out of labor force 0.02*** 0.02*** 0.02*** 0.03*** Wks unemployed x female 0.01* 0.003 0.003 0.001 Wks out of labor force x female 0.01*** 0.01*** 0.01*** 0.01** Health limitations 0.06 0.05 0.07 0.12* Health limitations x female 0.06 0.04 0.08 0.11 Number of children 0.05 ** 0.05*** 0.03 0.02 Number of children x female 0.13*** 0.15*** 0.11*** 0.1 0 Never married b 0.07 0.07 0.14 0.17 Married b 0.28 0.1 0.32 0.26 Separated b 0.34 0.06 0.34 0.31 Divorced b 0.15 0.08 0.18 0.19 Never married x female 0.07 c 0.02 0.06 Married x female 0.22 c 0.33 0.15 Separated x female 0.32 c 0.39 0.15 Divorced x female 0.01 c 0.11 c Females 563 529 435 216 Males 479 429 351 173 Notes: Year fixed effects also included but not reported. Significant at p<.1; ** Significant at p<.05; *** Significant at p<.01 a Reference category is near completers who are not enrolled, have not finished a degree, and do not enroll the following year. b Reference category is widowed. C Omitted due to multicollinearity.

PAGE 172

161 Table 29 : Differential effects by race/ethnicity Variable Def. 1 Def. 2 Def. 3 Def. 4 Pre enrollment 0.09* 0.09** 0.09* 0.04 Enrolled 0.12** 0.1 0 *** 0.04 0.03 Graduate 0.27*** 0.2 0 *** 0.26*** 0.21*** Pre enrollment x other ethnicity 0.04 0.04 0.12 0.18* Enrolled x other ethnicity 0.09* 0.09 0.15* 0.23** Graduate x other ethnicity 0.2 0 ** 0.1 0 0.26** 0.24** Age 0.11*** 0.07* 0.1** 0.11** Age 2 0.002*** 0.001 0.001 0.002*** Age x other ethnicity 0.01 0.04 0.002 0.035 Age 2 x other ethnicity 0.0002 0.0007 0.0001 0.0004 Exp 0.31*** 0.23*** 0.28*** 0.29*** Exp 2 0.03*** 0.02** 0.02*** 0.03*** Exp 3 0.002*** 0.001 0.001** 0.001** Exp 4 0.00003*** 0.00001 0.00002* 0.00002* Exp x other ethnicity 0.04 0.01 0.07 0.07 Exp 2 x other ethnicity 0.01 0.001 0.01 0.01 Exp 3 x other ethnicity 0.0004 0.0001 0.001 0.0005 Exp 4 x other ethnicity 0.00001 0.000001 0.00002 0.00001 Wks unemployed 0.03*** 0.03*** 0.03*** 0.02*** Wks out of labor force 0.02*** 0.02*** 0.02*** 0.03*** Wks unemployed x other ethnicity 0.003 0.003 0.003 0.005 Wks out of labor force x other ethnicity 0.01** 0.01 0.01* 0.01 Health limitations 0.02 0.05 0.02* 0.01 Health limitations x other ethnicity 0.01 0.04 0.01 0.1 Number of children 0.003 0.017 0.003 0.003 Number of children x other ethnicity 0.04 0.01 0.04 0.05 Never married b 0.05 0.06 0.08 0.02 Married b 0.12 0.05 0.1 0.08 Separated b 0.04 0.03 0.01 0.01 Divorced b 0.1 0.01 0.07 0.04 Never married x other ethnicity 0.08 0.1 0.03 0.40 Married x other ethnicity 0.07 0.15 0.04 0.45 Separated x other ethnicity 0.26 0.27 0.2 0.41 Divorced x other ethnicity 0.08 0.25 0.09 0.41 African Americans/Hispanics 477 457 391 185 Other Ethnicities 565 501 395 204 Notes: Year fixed effects also included but not reported. Significant at p<.1; ** Significant at p<.05; *** Significant at p<.01 a Reference category is near completers who are not enrolled, have not finished a degree, and do not enroll the following year. b Reference category is widowed.

PAGE 173

162 Table 30 : Differential effects by familial poverty status Variable Def. 1 Def. 2 Def. 3 Def. 4 Pre enrollment 0.09** 0.09*** 0.05 0.05 Enrolled 0.15*** 0.12*** 0.10** 0.05 Graduate 0.14*** 0.14** 0.09 0.11* Pre enrollment x poverty 0.12 0.06 0.01 0.07 Enrolled x poverty 0.02 0.02 0.05 0.003 Graduate x poverty 0.15 0.05 0.07 0.05 Age 0.12*** 0.06* 0.09*** 0.13** Age 2 0.001*** 0.001 0.001 0.002** Age x poverty 0.04 0.0 5 0.005 0.02 Age 2 x poverty 0.001 0.0003 0.0002 0.0003 Exp 0.27*** 0.26*** 0.25*** 0.33*** Exp 2 0.03*** 0.02*** 0.02** 0.03*** Exp 3 0.001*** 0.001** 0.001* 0.002 Exp 4 0.00003*** 0.00002* 0.00001 0.00003 Exp x poverty 0.05 0.12 0.06 0.06 Exp 2 x poverty 0.01 0.012 0.01 0.001 Exp 3 x poverty 0.0004 0.00 1 0.00 1 0.00001 Exp 4 x poverty 0.00001 0.0000 1 0.00001 0.000003 Wks unemployed 0.03*** 0.03*** 0.03*** 0.03*** Wks out of labor force 0.03*** 0.03*** 0.03*** 0.03*** Wks unemployed x poverty 0.004 0.001 0.004 0.004 Wks out of labor force x poverty 0.002 0.004 0.005 0.001 Health limitations 0.03 0.05 0.03 0.06 Health limitations x poverty 0.04 0.07 0.03 0.08 Number of children 0.03 0.0 3 0.04 0.02 Number of children x poverty 0.04 0.02 0.04 0.02 Never married b 0.11 0.01 0.07 0.1 Married b 0.19* 0.02 0.11 0.08 Separated b 0.17 0.04 0.08 0.16 Divorced b 0.22* 0.05 0.14 0.2 Never married x poverty 0.09 0.23 0.26 0.35 Married x poverty 0.03 0.19 0.17 0.34 Separated x poverty 0.03 0.25 0.13 0.26 Divorced x poverty 0.09 0.08 c c African Americans/Hispanics 780 702 584 296 Individuals in poverty, 1979 208 205 163 72 Notes: Year fixed effects also included but not reported. Significant at p<.1; ** Significant at p<.05; *** Significant at p<.01 a Reference category is near completers who are not enrolled, have not finished a degree, and do not enroll the following year. b Reference category is widowed. C Omitted due to multicollinearity

PAGE 174

163 Table 31 : Income differences by sector of graduation Variable Def. 1 Def. 2 Def. 3 Def. 4 Year prior to enrollment a 0.08 0.15 0.17 0.04 Enrolled a 0.31 *** 0.16 0.1 0.11 Graduate a 0.13 0.19 0.19* 0.09 Yr. prior to enr. x non profit 0.08 0.01 0.22 0.14 Yr. prior to enr. x for profit 0.29 0.04 0.2 0 0.02 Enrollee x non profit 0.07 0.05 0.07 0.02 Enrollee x for profit 0.06 0 0.001 0.34 Graduate x non profit 0.18 0.26 0.27 0.16 Graduate x for profit 0.22 0.07 0.09 0.1 Age a 0.12*** 0.06* 0.09** 0.17*** Age 2 a 0.001*** 0*** 0.001 0.002*** Exp a 0.26*** 0.24*** 0.23*** 0.31*** Exp 2 a 0.02*** 0.02** 0.02** 0.03*** Exp 3 a 0.001*** 0.001* 0.001 0.002*** Exp 4 a 0.00002*** 0.00001** 0.00001 0.00002** Wks unemployed 0.03*** 0.03*** 0.03*** 0.02*** Wks out of labor force 0.03*** 0.03*** 0.03*** 0.03*** Wks unemployed x non profit 0.01 c c c Wks unemployed x for profit 0.06 c c c Wks out of labor force x non profit 0.004 0.008 0.005 0.014** Wks out of labor force x for profit 0.02*** c c 0.03*** Health limitations 0.01 0 .002 0.02 0 .00003 Health limitations x non profit 0.03 0.07 0.05 0.03 Health limitations x for profit 0.16 0.24** c c Number of children 0.02 0.03 0.03 0.03 Number of children x non profit 0.07 0.04 0.02 0.04 Number of children x for profit 0.02 0.01 0.004 0.02 Never married b c 0.07 0.11 0.1 0.07 Married b c 0.13* 0.13 0.11 0.04* Separated b c 0.11 0.08 0.09 0.08 Divorced b c 0.12 0.14 0.11 0.05 Public Graduates 197 124 101 114 Non profit Graduates 74 43 36 46 For Profit Graduates 13 11 11 5 Notes: Year fixed effects also included but not reported. The interaction terms combine the dichotomous variable for sector of graduation with enrollment/graduation status. Significant at p<.1; ** Significant at p<.05; *** Significant at p<.01 a Reference category is near completers who are not enrolled, have not finished a degree, and do not enroll the following year. b Interaction terms for age, experience, and marriage eliminated due to multicollinearity. For profit i nteractions for health limitations, and non profit & for profit terms for out of the labor force also omitted. c Reference category is widowed.

PAGE 175

164 Definitional impacts on survival analyses The survival analysis models are also dependent on the definition of near completer. The different definitions lead to different individuals attaining near completer status and as such, present different samples from which to calculate survivorship. I follow the same procedures as above by testing each of the models with the different definitions. The results in the tables that follow compare the results across definitions. Several of the results from the general model are robust across definitions as can be seen in Table 32 Under all four definitions ASVAB score shows strong statistical significance (p<.01). The coefficients are also substantively similar when rounding is taken into account. The coefficient for paternal education also shows statistical significance, with three of the four definitions resulting in p values of less than .01 and the fourth showing modest significance at p<.10. Again, the coefficients are reasonably close to one another. Results for age are sensitive to the definition, with the definition in use showing no statistical s ignificance, but two of the other definition showing coefficients that are significant (p<.05). The coefficient for the number of children also shows at least some statistical significance across all four definitions. The coefficients vary somewhat but are consistently between .2 and .3 log points. The results for income vary across definitions. The results across the first two definitions show similar statistical significance and effect size, but the other two definitions result in coefficients that are no t statistically significant. It appears that as the sample shrinks the effect becomes insignificant. It could be that with larger samples of individuals meeting the criteria for the latter two definitions, the effect would become similar to that of the oth er two definitions. However, it could also be that the subtle changes in the sample criteria lead to substantively different results. Considering Definition 4, it could be that for these individuals who by definition tend to be closer to a degree incom e plays less of a role because they see it will take relatively little time to finish. With Definition 3, it could be that income plays a category). As individuals are awa y from school for longer periods, it could be that income becomes less of a determining factor. Given the smaller sample sizes, however, it is prudent to refrain from drawing firm conclusions here, although further research is certainly warranted. Tables 33 35 show the comparisons for the event history analyses with interaction effects. Generally speaking, relatively few interacted terms show statistical significance across any of the models. For the model interacting gender terms, Definition 4 shows a sta tistically significant term for female age, experience, and number of children (though the later is only significant at p<.10). This suggests that there are some differences across genders for this sample, which tends to be closer to a degree at the time w hen stop out occurred. In the model with interaction terms for racial/ethnic group, which is presented in T able 34 only the interaction term between other ethnicities and receipt of newspapers in 1979 shows statistical significance in any of the models. This term is only significant when using Definitions 1 and 2. The large effect size across these two definitions, but the absence of statistical significance under the other two definitions suggests that this result should be evaluated further, perhaps by considering other proxies for the educational environment in which an individual is raised.

PAGE 176

165 None of the interaction terms in Table 35 which presents a model with interactions for family poverty status at the outset of the survey, are statistically signifi cant. This lends some weight to the conclusion that this characteristic becomes less important in predicting educational attainment once individuals reach the threshold of having completed significant college credit. Table 32 : Even t history analysis using different near completer definitions Variable Def. 1 Def. 2 Def. 3 Def. 4 Time Invariant Characteristics Maternal Education 0.005 0.04 0.01 0.01 Paternal Education 0.08*** 0.05* 0.08** 0.08*** Poverty Status (1979) 0.22 0.09 0.6** 0.17 Library Card (1979) 0.24 0.09 0.25 0.23 HH Rec'd Newspaper (1979) 0.13 0.04 0.22 0.15 HH Rec'd Magazines (1979) 0.12 0.15 0.01 0.17 Fem ale 0.02 0.01 0.28 0.2 0 ASVAB percentile 0.02*** 0.01*** 0.02*** 0.01*** Other Ethnicity 0.03 0.05 0.06 0.03 Time Variant Characteristics Former military status 0.18 0.4 0 0.33 0.01 Age 0.04 0.05** 0.03 0.06** Experience 0.04 0.04 0.09** 0.01 Number of Children 0.2 0 0.27*** 0.29** 0.2 0 Health Status 0.03 0.28 0.25 0.11 Log Income 0.23** 0.24** 0.04 0.04 Out of Labor Force 0.01* 0.01 0.01 0.004 Unemployed 0.03* 0.01 0.01 0.004 Subjects 952 886 741 350 Failures 241 205 119 145 Notes: **Significant at p<.01; **Significant at p<.05, *Significant at p< .10. Time variant variables lagged prior to final enrollment spell

PAGE 177

166 Table 33 : Event history analysis with gender interaction terms Variable Def. 1 Def. 2 Def. 3 Def. 4 Time Invariant Characteristics Maternal Education 0.03 0.01 0.07 0.01 Maternal Education x female 0.04 0.08 0.09 0.02 Paternal Education 0.07** 0.07** 0.05 0.06* Paternal Education x female 0.02 0.07 0.05 0.04 Poverty Status (1979) 0.05 0.04 0.12* 0.03 Poverty Status (1979) x female 0.29 0.19 1 .00 0.17 Library Card (1979) 0.11 0.14 0.24 0.24 Library Card (1979) x female 0.15 0.26 0.16 0.37 HH Rec'd Newspaper (1979) 0.13 0.15 0.2 0 0.05 HH Rec'd Newspaper (1979) x female 0.49 0.19 0.04 0.47 HH Rec'd Magazines (1979) 0.46* 0.26 0.17 0.11 HH Rec'd Magazines (1979) x female 0.71* 0.2 0 0.32 0.08 Female 1.25 a 0.99 1.08 ASVAB percentile 0.03*** 0.02*** 0.03*** 0.02*** ASVAB percentile x female 0.01 0.01* 0.01 0.01 Other Ethnicity 0.07 0.07 0.06 0.02 Other Ethnicity x female 0.21 0.003 0.02 0.11 Time Variant Characteristics Former military status 0.1 0 0.32 0.24 0.001 Former military status x female 0.38 0.02 0.18 0.42 Age 0.04 0.07 0.01 0.01 Age x female 0.004 0.03 0.07 0.11** Experience 0.1 0 ** 0.11** 0.13** 0.12*** Experience x female 0.09 0.09 0.04 0.23*** Number of Children 0.07 0.04 0 .004 0.05 Number of Children x female 0.2 0 0.45 0.5 0.42* Health Status 0.19 0.56 0.48 0.63 Health Status x female 0.28 0.39 0.26 0.58 Log Income 0.33** 0.6 0 *** 0.06 0.18 Log Income x female 0.18 0.65*** 0.07 0.24 Out of Labor Force 0.02 0.02 0 .001 0.004 Out of Labor Force x female 0.01 0.002 0.01 0.02 Unemployed 0.05** 0.03* 0.02 0.01 Unemployed x female 0.04 0.05 0.01 0.02 Male events: 118 97 52 79 Female events: 123 108 67 66 Notes: **Significant at p<.01; **Significant at p<.05, *Significant at p<.10. Time variant characteristics lagged prior to final pre enrollment spell. a Omitted due to multicollinearity errors.

PAGE 178

167 Table 34 : Event history analysis with racial/ethnic interaction terms Variable Def. 1 Def. 2 Def. 3 Def. 4 Time Invariant Characteristics Maternal Education 0.05 0.07* 0.04 0.02 Maternal Education x other ethnicity 0.08 0.08 0.04 0.08 Paternal Education 0.05 0.04 0.12** 0.05 Paternal Education x other ethnicity 0.04 0.02 0.07 0.05 Poverty Status (1979) 0.39 0.09 0.89** 0.25 Poverty Status (1979) x other ethnicity 0.22 0.1 0 0.66 0.08 Library Card (1979) 0.09 0.22 0.51 0.003 Library Card (1979) x other ethnicity 0.36 0.23 0.53 0.7 0 HH Rec'd Newspaper (1979) 0.56* 0.38 0.04 0.27 HH Rec'd Newspaper (1979) x other ethnicity 1.24** 1.07** 0.19 1.1 0 HH Rec'd Magazines (1979) 0.14 0.17 0.14 0.09 HH Rec'd Magazines (1979) x other ethnicity 0.08 0.15 0.24 0.54 Female 0.28 0.17 0.46 0.13 Female x other ethnicity 0.45 0.3 0 0.25 0.14 ASVAB percentile 0.02*** 0.01** 0.02*** 0.02*** ASVAB percentile x other ethnicity 0.002 0.003 0.004 0.01 Other Ethnicity 0.93 2.96 0.51 0.59 Time Variant Characteristics Former military status 0.07 0.58 0.48 0.21 Former military status x other ethnicity 0.13 0.32 0.33 0.24 Age 0.03 0.02 0.07 0.04 Age x other ethnicity 0.01 0.06 0.07 0.05 Experience 0.06 0.09** 0.14*** 0.04 Experience x other ethnicity 0.04 0.1 0 0.1 0 0.07 Number of Children 0.1 0 0.28** 0.38 0.11 Number of Children x other ethnicity 0.25 0.02 0.1 0 0.22 Health Status 0.23 0.3 0 0.5 0 0.27 Health Status x other ethnicity 0.34 0.05 0.35 0.29 Log Income 0.24 0.15 0.11 0.08 Log Income x other ethnicity 0.004 0.21 0.24 0.06 Out of Labor Force 0.02 0.01 0.01 0.005 Out of Labor Force x other ethnicity 0.01 0.004 0.01 0.02 Unemployed 0.04* 0.01 0.01 0.01 Unemployed x other ethnicity 0.02 0.01 0.02 0.02 African American/Hispanic events 89 92 52 59 Other Ethnicities events 152 113 45 86 Notes: **Significant at p<.01; **Significant at p<.05, *Significant at p<.10. Time variant characteristics lagged prior to final pre enrollment spell.

PAGE 179

168 Table 35 : Event history analysis sensitivity tests with poverty status interactions Variable Def. 1 Def. 2 Def. 3 Def. 4 Time Invariant Characteristics Maternal Education 0.004 0.04 0.02 0.01 Maternal Education x poverty status (1979) 0.02 0.004 0.11 0.003 Paternal Education 0.09*** 0.05* 0.09** 0.08*** Paternal Education x poverty status (1979) 0.09 0.02 0.05 0.06 Poverty Status (1979) 1.74 0.7 0 7.06 1.29 Library Card (1979) 0.19 0.07 0.26 0.04 Library Card (1979) x poverty status (1979) 0.23 0.63 0.55 0.65 HH Rec'd Newspaper (1979) 0.16 0.04 0.12 0.22 HH Rec'd Newspaper (1979) x poverty status (1979) 0.24 0.02 1.13 0.32 HH Rec'd Magazines (1979) 0.11 0.18 0.06 0.13 HH Rec'd Magazines (1979) x poverty status (1979) 0.0003 0.16 0.71 0.16 Female 0.08 0.13 0.41 0.03 Female x poverty status (1979) 0.43 0.63 1 .00 1.21** ASVAB percentile 0.02*** 0.01*** 0.02*** 0.01*** ASVAB percentile x poverty status (1979) 0.001 0.002 0.03 0.01 Other ethnicity 0.05 0.08 0.06 0.1 0 Other ethnicity x poverty status (1979) 0.17 0.18 0.15 0.5 0 Time Variant Characteristics Former military status 0.01 0.19 0.01 0.3 0 Former military status x poverty status (1979) 0.86 1.08 9.62 0.96 Age 0.05* 0.06** 0.02 0.08*** Age x poverty status (1979) 0.06 0.04 0.13 0.1 0 Experience 0.03 0.04 0.07* 0.02 Experience x poverty status (1979) 0.04 0.01 0.21 0.07 Number of Children 0.17 0.29*** 0.27** 0.21* Number of Children x poverty status (1979) 0.13 0.11 0.15 0.14 Health Status 0.03 0.28 0.24 0.07 Health Status x poverty status (1979) 0.26 0.06 0.3 0 0.5 0 Log Income 0.2 0 ** 0.25** 0.03 0.0003 Log Income x poverty status (1979) 0.23 0.02 0.38 0.24 Out of Labor Force 0.01 0.01 0.001 0.01 Out of Labor Force x poverty status (1979) 0.01 0.001 0.08 0.01 Unemployed 0.02 0.003 0.0004 0.004 Unemployed x poverty status (1979) 0.06 0.02 0.05 0.04 Poverty events 89 38 13 19 Non poverty events 152 167 106 126 Notes: **Significant at p<.01; **Significant at p<.05, *Significant at p<.10. Time variant characteristics lagged prior to final pre enrollment spell.

PAGE 180

169 Wage Premiums over Time While the main fixed effects model show that there does appear to be a wage premium for degree completion, the coefficient effectively represents the average premium over all of the years in the survey after an individual graduates. Including additional va riables to represent the years since graduation can examine whether the premium for degree completion in constant over time, or whether it changes. This follows an approach used by Evans, Murray, and Schwab (1997). It could be that degree completers receiv e an initial wage benefit, then follow the same earnings trajectory as non completers. Alternatively, it could be that degree completion alters the earnings trajectory such that there is little immediate benefit, but it grows over time. Repeating the init ial analysis with a series of dichotomous variables representing years since graduation shows that the wage premium does appear to increase over time. The results are presented in Table 36 below. The coefficients for years since graduation Table 36 : Categorical years since graduation Variable Coefficient Std. Err Pre enrollment a 0.06 0.03 Enrolled a 0.18 *** 0.03 Graduate a 0.06 0.05 4 6 yrs since grad b 0.31 *** 0.04 7 9 yrs since grad b 0.39 *** 0.04 10 12 yrs since grad b 0.39 *** 0.06 13 15 yrs since grad b 0.4 0*** 0.06 16+ yrs since grad b 0.45 *** 0.07 Age 0.12 *** 0.03 Age 2 0.001 *** 0.0004 Exp 0.29 *** 0.04 Exp 2 0.03 *** 0.01 Exp 3 0.002 *** 0.0003 Exp 4 0.00003 *** 0.000007 Wks unemployed 0.03 *** 0.002 Wks out of labor force 0.03 *** 0.002 Health limitations 0.02 0.02 Number of children 0.02 0.02 Never married c 0.11 0.11 Married c 0.15 0.11 Separated c 0.13 0.12 Divorced c 0.14 0.11 Observations 12,156 1,042 Individuals Note: Year fixed effects also included but not reported. Significant at p<.1; ** Significant at p<.05; *** Significant at p<.01 a Reference category is near completers who are not enrolled, have not finished a degree, and do not enroll the following yea r. b Reference category is 1 3 years since graduation. c Reference category is widowed

PAGE 181

170 are all positive and statistically significant (p< .01). The reference category for these variables is 1 3 years since graduation. Testing the equivalence of the coefficients suggests that the wage premium does increase over time, as the coefficient for 16 or more years since graduation is significantly di fferent from the coefficient for 4 6 years since graduation. A joint test of all of these variables shows that they are jointly significant. An alternative approach is to use the years since graduation as a continuous independent variable. In Table 37 belo w, I also include a quadratic term, which is statistically significant and negative, suggesting that while the wage premium may increase over time, the rate of increase slows the longer a near completer is away from graduation. These results are still cons istent with the approach using dichotomous variables above and provide additional evidence that the wage premium is not constant over time. In this model, the coefficient for graduation is statistically significant (p<.05) and negative, which suggests that the income premium may still be negative immediately Table 37 : Continuous years since graduation Variable Coefficient Std. Err Pre enrollment a 0.06* 0.03 Enrolled a 0.18*** 0.03 Graduate a 0.11** 0.05 Years from graduation 0.06*** 0.01 Years from graduation 2 0.002*** 0.0003 Age 0.12*** 0.03 Age 2 0.001*** 0.0003 Exp 0.29*** 0.04 Exp 2 0.03*** 0.01 Exp 3 0.002*** 0.0003 Exp 4 0.00003*** 0.00001 Wks unemployed 0.03*** 0.002 Wks out of labor force 0.03*** 0.002 Health limitations 0.02 0.02 Number of children 0.02 0.02 Never married b 0.11 0.1 0 Married b 0.16 0.1 0 Separated b 0.13 0.11 Divorced b 0.15 0.1 0 Observations 12,156 1,042 Individuals Note: Year fixed effects also included but not reported. Significant at p<.1; ** Significant at p<.05; *** Significant at p<.01 a Reference category is near completers who are not enrolled, have not finished a degree, and do not enroll the following year. b Reference category is widowed

PAGE 182

171 following graduation. Taken collectively, Tables 36 and 37 demonstrate that the wage premium that baccalaureate degree completers earn is likely not constant and varies over time. These results bear consideration in future research. Including a variable wage premi um in the approximations for rate of return would result in lower overall premiums as the highly lucrative years would occur farther in the future (from the decision point of deciding whether or not to go back to finish a degree) and would thus be reduced more by the discount rate. Additional Scenarios for Rates of Return The scenarios approximating the rate of return for baccalaureate degree completion by near completers vary direct costs (based on average tuition rates by sector). These approximations do not, however, vary the indirect costs even though it is likely that individuals returning to finish degrees have varied enrollment intensity that affects their direct costs as well. Many individuals may choose to take fewer classes in order to work more, which presumably results in fewer foregone wages, but also perhaps more years to finish a degree. This section presents additional scenarios for the rates of return based on longer periods of reenrollment and lower indirect costs. Full indirect costs = 11% of wages, 2 yrs to graduate, 17% direct costs Half indirect costs = 5.5% of wages, 4 yrs to graduate, 8.5% direct costs One third indirect costs = 3.7% of wages, 6 yrs to graduate, 5.5% direct costs Figure 20 : African Ameri can/Hispanic rate of return with varied indirect costs Figure 20 shows additional approximations for the rate of return using the wage premium estimated for African Americans/Hispanics, which was 31 percent. In the three approximations presented in this f igure, as indirect costs decrease, I increase the length of enrollment to simulate what is likely to happen as an individual decreases their enrollment intensity. Intuitively this may interfere less with their employment, resulting in fewer foregone wages, but results in additional years of enrollment. Based on these approximations, the initial loss may be less for individuals foregoing fewer wages, but

PAGE 183

172 the longer term return is slightly less due to the additional years of enrollment. Here the discount rate comes into play as nearer term effects (particularly lower wages during the longer enrollment period) affect the longer term wage gains more than higher earnings far in the future. Based on these rough approximations, it appears from a purely economic per spective to make more sense for these individuals to increase enrollment intensity to earnings. Of course, reality may not follow these approximations exactly. Weighted Re gression I elect not to employ survey weights in the main models presented in the body of this research While there is some justification for this decision from the econometric literature, it can still be instructive to compare results for weighted an d non weighted regressions. Table 3 8 presents results for the two regression models. These results show Table 38 : Comparison of weighted and non weighted regressions Variable Non weighted Weighted Year prior to enrollment a 0.07** 0.04 Enrolled a 0.17*** 0.19*** Graduate a 0.16*** 0.09** Age 0.12*** 0.09*** Age 2 0.001*** 0.001*** Exp 0.29*** 0.29*** Exp 2 0.03*** 0.03*** Exp 3 0.001*** 0.002*** Exp 4 0.00002*** 0.00003*** Wks unemployed 0.03*** 0.03*** Wks out of labor force 0.03*** 0.03*** Health limitations 0.02 0.01 Number of children 0.02 0.02 Never married b 0.08 0.02 Married b 0.14 0.09 Separated b 0.12 0.11 Divorced b 0.13 0.07 Note: Year fixed effects also included but not reported. Significant at p<.1; ** Significant at p<.05; *** Significant at p<.01 a Reference category is near completers who are not enrolled, have not finished a degree, and do not enroll the following year. b Reference category is widowed that the weighted regression produces a smaller coefficient for graduates. Additionally, the coefficient for pre enrollment is not statistically significant in the weighted model. The rest of the coefficients are comparable. It is not possible to tell, however whether the

PAGE 184

173 difference between the two coefficients is statistically significant because they come from different regression models. The 95 percent confidence bounds for the two coefficients overlap, so it may be that these two outcomes are not substanti vely different.